Patent Grant
7428003
1. Field of the Invention
The present invention relates to an automatic stabilization control apparatus, an automatic stabilization control method, and a recording medium having an automatic stabilization control program recorded thereon.
2. Description of the Background Art
Automatic stabilization control apparatuses for carrying out automatic stabilization control of an image picked up by an image pick-up device such as a video camera have been conventionally known.
The geometric conversion coefficient calculation unit 51 calculates a geometric conversion coefficient utilizing image processing such as matching from a plurality of frames, for example. The automatic stabilization control unit 52 subjects an input frame which is stored in a memory once to geometric conversion using the geometric conversion coefficient calculated by the geometric conversion coefficient calculation unit 51.
The geometric conversion coefficient is represented by a matrix of 3 by 3, as expressed by the following equation (1) and corresponds to a coefficient in world coordinate conversion between two frames.
Let FS be a start frame (a reference frame), F2 be the current input frame, F1 be an input frame which is one frame preceding the current input frame F2, M12 be a geometric conversion coefficient between the frames F1 and F2 (a geometric conversion coefficient for converting coordinates on the frame F2 into coordinates on the frame F1), MS1 be a geometric conversion coefficient between the frames FS and F1, and MS2 be an accumulated geometric conversion coefficient between the frames FS and F2 (an accumulated geometric conversion coefficient for converting coordinates on the frame F2 into coordinates on the frame FS).
Each of the frames from the start frame to the end frame is subjected to the following processing: That is, an output frame memory is cleared (step 100), the frame is read on a work memory, and the read frame (input frame) is taken as F2 (step 101).
It is then judged whether or not the input frame F2 is the start frame FS (step 102). When it is judged that the input frame F2 is the start frame FS, a coefficient (an initial value) representing non-conversion as expressed by the following equation (2) is set as the accumulated geometric conversion coefficient MS2 between the start frame FS and the input frame F2 (step 103).
The input frame F2 is set as the preceding input frame F1, and MS2 is set as the geometric conversion coefficient MS1 between the start frame FS and the preceding input frame F1 (step 104). Further, data representing the input frame F2 is rendered as it is in the output frame memory (step 105). Thereafter, the procedure is returned to the step 100. In the step 100, the output frame memory is cleared, and the succeeding frame is read.
When it is judged in the foregoing step 102 that the input frame F2 is not the start frame FS, the geometric conversion coefficient M12 between the preceding frame F1 and the current frame F2 is calculated (step 106). The accumulated geometric conversion coefficient MS2 between the start frame FS and the current frame F2 is then calculated from MS1 and M12, as expressed by the following equation (3) (step 107)
MS2=M12MS1 (3)
The input frame F2 is set as the preceding input frame F1, and MS2 is set as the geometric conversion coefficient MS1 between the start frame FS and the preceding input frame F1 (step 108). Further, the input frame F2 is subjected to geometric conversion using the accumulated geometric conversion coefficient MS2, and the frame after the geometric conversion is then rendered in the output frame memory (step 109). Thereafter, the procedure is returned to the step 100. In the step 100, the output frame memory is cleared, and the succeeding frame is read.
A typical method of calculating a geometric conversion coefficient between two frames will be described (see JP-A-11-339021). First, a feature point such as an edge is extracted from the one frame. Feature points are then tracked (matched) between the frames, to find a correspondence point on the one frame corresponding to the feature point on the other frame. Two equations such as an equation (4) are obtained from one set of correspondence points (x1, y1) and (x1′, y1′), thereby deriving six or more equations by detecting three or more sets of correspondence points. From the equations, a geometric conversion coefficient with unknowns is calculated using a linear solution.
x′=m11x+m12y+m13 y′=m21x+m22y+m23 (4)
In the conventional geometric conversion coefficient calculating method, the accumulated geometric conversion coefficient between the start frame FS and the input frame F2 is calculated by accumulating the geometric conversion coefficients between the adjacent frames from the start frame FS to the input frame F2. Therefore, as the number of input frames increases, calculation errors of the geometric conversion coefficients between the adjacent frames are accumulated. That is, while a large error is made in calculating the geometric conversion coefficient in the frame intermediate between the start frame and the input frame, the frames succeeding the frame corrected by the erroneous conversion coefficient are corrected.
Furthermore, in the conventional automatic stabilization control method, an area where no rendering is performed (hereinafter referred to as a non-rendering area) occurs in the output frame memory. That is, when a plurality of frames picked up in a state where hands move are frames as shown in
An object of the present invention is to provide an automatic stabilization control apparatus capable of finding a geometric conversion coefficient between a reference frame and an input frame with high precision and consequently, allowing high-precision automatic stabilization control to be carried out, an automatic stabilization control method, and a recording medium having an automatic stabilization control program recorded thereon.
Another object of the present invention is to provide an automatic stabilization control apparatus capable of preventing, in a memory area for an output frame storing an image after automatic stabilization control, a non-rendering area from occurring, an automatic stabilization control method, and a recording medium having an automatic stabilization control program recorded thereon.
A further object of the present invention is to provide an automatic stabilization control apparatus capable of preventing, in a memory area for an output frame storing an image after automatic stabilization control, a non-rendering area from occurring, and reducing degradation in the quality of the image with its angle of field held, an automatic stabilization control method, and a recording medium having an automatic stabilization control program recorded thereon.
In an automatic stabilization control apparatus that makes automatic stabilization control of data representing a moving image within a range from a reference frame to a frame which is an arbitrary number of frames succeeding the reference frame, a first automatic stabilization control apparatus according to the present invention is characterized by comprising first means for calculating a geometric conversion coefficient between an input frame and a frame which is one frame preceding the input frame for converting coordinates on the input frame into coordinates on the frame which is one frame preceding the input frame as a geometric conversion coefficient between the adjacent frames; second means for accumulating the geometric conversion coefficients between the adjacent frames from the reference frame to the input frame, to calculate an indirect geometric conversion coefficient between the input frame and the reference frame; third means for calculating a direct geometric conversion coefficient between the input frame and the reference frame utilizing the indirect geometric conversion coefficient between the input frame and the reference frame; and fourth means for carrying out automatic stabilization control of the input frame using the direct geometric conversion coefficient between the input frame and the reference frame.
An example of the third means is one comprising means for finding a plurality of feature points from the input frame, means for finding a correspondence point on the reference frame for each of the feature points, and means for calculating the direct geometric conversion coefficient between the input frame and the reference frame on the basis of each of the feature points on the input frame and the correspondence point on the reference frame for the feature point, the means for finding the correspondence point on the reference frame for each of the feature points converting the coordinates of the feature point on the input frame into coordinates on the reference frame using the indirect geometric conversion coefficient between the input frame and the reference frame, and tracking the correspondence point on the reference frame corresponding to the feature point on the input frame using an obtained coordinate point on the reference frame as an initial value, to find the correspondence point on the reference frame corresponding to the feature point on the input frame.
An example of the third means is one comprising means for finding a plurality of feature points from the reference frame, means for finding a correspondence point on the input frame for each of the feature points, and means for calculating a direct geometric conversion coefficient between the input frame and the reference frame on the basis of each of the feature points on the reference frame and the correspondence point on the input frame for the feature point, the means for finding the correspondence point on the input frame for each of the feature points converting the coordinates of the feature point on the reference frame into coordinates on the input frame using a coefficient obtained by subjecting the indirect geometric conversion coefficient between the input frame and the reference frame to inverse conversion, and tracking the correspondence point on the input frame corresponding to the feature point on the reference frame using an obtained coordinate point on the input frame as an initial value, to find the correspondence point on the input frame corresponding to the feature point on the reference frame.
In an automatic stabilization control method for carrying out automatic stabilization control of data representing a moving image within a range from a reference frame to a frame which is an arbitrary number of frames succeeding the reference frame, a first automatic stabilization control method according to the present invention is characterized by comprising a first step of calculating a geometric conversion coefficient between an input frame and a frame which is one frame preceding the input frame for converting coordinates on the input frame into coordinates on the frame which is one frame preceding the input frame as a geometric conversion coefficient between the adjacent frames; a second step of accumulating the geometric conversion coefficients between the adjacent frames from the reference frame to the input frame, to calculate an indirect geometric conversion coefficient between the input frame and the reference frame; a third step of calculating a direct geometric conversion coefficient between the input frame and the reference frame utilizing the indirect geometric conversion coefficient between the input frame and the reference frame; and a fourth step of carrying out automatic stabilization control of the input frame using the direct geometric conversion coefficient between the input frame and the reference frame.
An example of the third step is one comprising the steps of finding a plurality of feature points from the input frame, finding a correspondence point on the reference frame for each of the feature points, and calculating the direct geometric conversion coefficient between the input frame and the reference frame on the basis of each of the feature points on the input frame and the correspondence point on the reference frame for the feature point, the step of finding the correspondence point on the reference frame for each of the feature points comprising the step of converting the coordinates of the feature point on the input frame into coordinates on the reference frame using the indirect geometric conversion coefficient between the input frame and the reference frame, and tracking the correspondence point on the reference frame corresponding to the feature point on the input frame using an obtained coordinate point on the reference frame as an initial value, to find the correspondence point on the reference frame corresponding to the feature point on the input frame.
An example of the third step is one comprising the steps of finding a plurality of feature points from the reference frame, finding a correspondence point on the input frame for each of the feature points, and calculating the direct geometric conversion coefficient between the input frame and the reference frame on the basis of each of the feature points on the reference frame and the correspondence point on the input frame for the feature point, the step of finding the correspondence point on the input frame for each of the feature points comprising the step of converting the coordinates of the feature point on the reference frame into coordinates on the input frame using a coefficient obtained by subjecting the indirect geometric conversion coefficient between the input frame and the reference frame to inverse conversion, and tracking the correspondence point on the input frame corresponding to the feature point on the reference frame using an obtained coordinate point on the input frame as an initial value, to find the correspondence point on the input frame corresponding to the feature point on the reference frame.
In a computer readable recording medium having an automatic stabilization control program, for carrying out automatic stabilization control of data representing a moving image within a range from a reference frame to a frame which is an arbitrary number of frames succeeding the reference frame, recorded thereon, a first recording medium according to the present invention is characterized in that the automatic stabilization control program causes a computer to carry out a first step of calculating a geometric conversion coefficient between an input frame and a frame which is one frame preceding the input frame for converting coordinates on the input frame into coordinates on the frame which is one frame preceding the input frame as a geometric conversion coefficient between the adjacent frames; a second step of accumulating the geometric conversion coefficients between the adjacent frames from the reference frame to the input frame, to calculate an indirect geometric conversion coefficient between the input frame and the reference frame; a third step of calculating a direct geometric conversion coefficient between the input frame and the reference frame utilizing the indirect geometric conversion coefficient between the input frame and the reference frame; and a fourth step of carrying out automatic stabilization control of the input frame using the direct geometric conversion coefficient between the input frame and the reference frame.
An example of the third step is one comprising the steps of finding a plurality of feature points from the input frame, finding a correspondence point on the reference frame for each of the feature points, and calculating the direct geometric conversion coefficient between the input frame and the reference frame on the basis of each of the feature points on the input frame and the correspondence point on the reference frame for the feature point, the step of finding the correspondence point on the reference frame for each of the feature points comprising the step of converting the coordinates of the feature point on the input frame into coordinates on the reference frame using the indirect geometric conversion coefficient between the input frame and the reference frame, and tracking the correspondence point on the reference frame corresponding to the feature point on the input frame using an obtained coordinate point on the reference frame as an initial value, to find the correspondence point on the reference frame corresponding to the feature point on the input frame.
An example of the third step is one comprising the steps of finding a plurality of feature points from the reference frame, finding a correspondence point on the input frame for each of the feature points, and calculating the direct geometric conversion coefficient between the input frame and the reference frame on the basis of each of the feature points on the reference frame and the correspondence point on the input frame for the feature point, the step of finding the correspondence point on the input frame for each of the feature points comprising the step of converting the coordinates of the feature point on the reference frame into coordinates on the input frame using a coefficient obtained by subjecting the indirect geometric conversion coefficient between the input frame and the reference frame to inverse conversion, and tracking the correspondence point on the input frame corresponding to the feature point on the reference frame using an obtained coordinate point on the input frame as an initial value, to find the correspondence point on the input frame corresponding to the feature point on the reference frame.
In an automatic stabilization control apparatus that makes automatic stabilization control of data representing a moving image, a second automatic stabilization control apparatus according to the present invention is characterized by comprising first means for calculating for each of input frames a geometric conversion coefficient between a frame to be the basis of automatic stabilization control and the input frame; and second means for producing for each of the input frames an output frame, the second means comprising means for clearing an output frame memory, means for subjecting the input frame to geometric conversion on the basis of the geometric conversion coefficient, corresponding to the input frame, calculated by the first means, means for rendering a frame, after the geometric conversion, corresponding to the input frame in the output frame memory, and means for interpolating a non-rendering area within the output frame memory by at least one frame in the vicinity of the input frame.
In an automatic stabilization control method for carrying out automatic stabilization control of data representing a moving image, a second automatic stabilization control method according to the present invention is characterized by comprising a first step of calculating for each of input frames a geometric conversion coefficient between a frame to be the basis of automatic stabilization control and the input frame; and a second step of producing for each of the input frames an output frame, the second step comprising the steps of clearing an output frame memory, subjecting the input frame to geometric conversion on the basis of the geometric conversion coefficient, corresponding to the input frame, calculated in the first step, rendering a frame, after the geometric conversion, corresponding to the input frame in the output frame memory, and interpolating a non-rendering area within the output frame memory by at least one frame in the vicinity of the input frame.
In a computer readable recording medium having an automatic stabilization control program, for carrying out automatic stabilization control of data representing a moving image, recorded thereon, a second recording medium according to the present invention is characterized in that the automatic stabilization control program causes a computer to carry out a first step of calculating for each of input frames a geometric conversion coefficient between a frame to be the basis of automatic stabilization control and the input frame; and a second step of producing for each of the input frames an output frame, the second step comprising the steps of clearing an output frame memory, subjecting the input frame to geometric conversion on the basis of the geometric conversion coefficient, corresponding to the input frame, calculated in the first step, rendering a frame, after the geometric conversion, corresponding to the input frame in the output frame memory, and interpolating a non-rendering area within the output frame memory by at least one frame in the vicinity of the input frame.
In an automatic stabilization control apparatus that makes automatic stabilization control of data representing a moving image, a third automatic stabilization control apparatus according to the present invention is characterized by comprising first means for calculating for each of input frames a geometric conversion coefficient between a frame to be the basis of automatic stabilization control and the input frame; and second means for subjecting for each of the input frames the input frame to geometric conversion on the basis of the geometric conversion coefficient, corresponding to the input frame, calculated by the first means, and then rendering a frame, after the geometric conversion, corresponding to the input frame in the output frame memory; and third means for holding the results of the rendering by the second means, the second means overwriting, in rendering frames, after the geometric conversion, corresponding to the second and the subsequent input frames in the output frame memory, the frames after the geometric conversion corresponding to the input frames on the results of the rendering so far performed.
In an automatic stabilization control method for carrying out automatic stabilization control of data representing a moving image, a third automatic stabilization control method according to the present invention is characterized by comprising a first step of calculating for each of input frames a geometric conversion coefficient between a frame to be the basis of automatic stabilization control and the input frame; a second step of subjecting for each of the input frames the input frame to geometric conversion on the basis of the geometric conversion coefficient, corresponding to the input frame, calculated in the first step, and then rendering a frame, after the geometric conversion, corresponding to the input frame in the output frame memory; and a third step of holding the results of the rendering in the second step, the second step comprising the step of overwriting, in rendering frames, after the geometric conversion, corresponding to the second and the subsequent input frames in the output frame memory, the frames after the geometric conversion corresponding to the input frames on the results of the rendering so far performed.
In a computer readable recording medium having an automatic stabilization control program, for carrying out automatic stabilization control of data representing a moving image, recorded thereon, a third recording medium according to the present invention is characterized in that the automatic stabilization control program causes a computer to carry out a first step of calculating for each of input frames a geometric conversion coefficient between a frame to be the basis of automatic stabilization control and the input frame; a second step of subjecting for each of the input frames the input frame to geometric conversion on the basis of the geometric conversion coefficient, corresponding to the input frame, calculated in the first step, and then rendering a frame, after the geometric conversion, corresponding to the input frame in the output frame memory; and a third step of holding the results of the rendering in the second step, the second step comprising the step of overwriting, in rendering frames, after the geometric conversion, corresponding to the second and the subsequent input frames in the output frame memory, the frames after the geometric conversion corresponding to the input frames on the results of the rendering so far performed.
The foregoing and other objects, features, aspects and advantages of the present invention will become more apparent from the following detailed description of the present invention when taken in conjunction with the accompanying drawings.
Embodiments of the present invention will be described.
[1] Description of Automatic Stabilization Control Apparatus
The automatic stabilization control apparatus is realized by a personal computer (PC) 10. A display 21, a mouse 22, and a keyboard 23 are connected to the PC 10. The PC 10 comprises a CPU 11, a memory 12, a hard disk 13, and a drive of a removable disk such as a CD-ROM 14 (a disk drive).
The hard disk 13 stores an automatic stabilization control program in addition to an OS (Operation System) or the like. The automatic stabilization control program is installed in the hard disk 13 using a CD-ROM 20 storing the program. The hard disk 13 previously stores a moving image file picked up by a video camera or the like.
[2] Description of Automatic Stabilization Control Processing Performed by CPU in Case where Automatic Stabilization Control Program is Started
Examples of automatic stabilization control processing include geometric conversion coefficient calculation processing and output frame production processing. First, the output frame production processing is performed after the geometric conversion coefficient calculation processing is performed.
In the present embodiment, a user is caused to previously set judgment whether or not a non-rendering area in an output frame memory is subjected to interpolation processing. In a case where such setting as to subject the non-rendering area to interpolation processing is performed, there occurs a state where a flag storing the fact (hereinafter referred to as a flag for judging whether or not interpolation is required) flag is set (flag=1). On the other hand, in a case where such setting as to subject the non-rendering area to interpolation processing is not performed, there occurs a state where the flag for judging whether or not interpolation is required flag is reset (flag=0).
In the automatic stabilization control processing, the first frame (a start frame) in a range where automatic stabilization control is carried out is used as a reference frame, to carry out automatic stabilization control of the other frame in the range where automatic stabilization control is carried out. The range in which automatic stabilization control is carried out can be arbitrarily set.
Let FS be a start frame (a reference frame) in the range where automatic stabilization control is carried out, F2 be the current input frame in the range where automatic stabilization control is carried out, F1 be an input frame which is one frame preceding the current input frame F2, M12 be a geometric conversion coefficient between the frames F1 and F2 (a geometric conversion coefficient for converting coordinates on the frame F2 into coordinates on the frame F1) MS1 be a geometric conversion coefficient between the frames FS and F1, and MS2 be an accumulated geometric conversion coefficient between the frames FS and F2 (an accumulated geometric conversion coefficient for converting coordinates on the frame F2 into coordinates on the frame FS).
Each of the frames from the start frame to the end frame in the range where automatic stabilization control is carried out is subjected to the following processing: That is, the frame is first read onto a memory, and the read frame (input frame) is taken as F2 (step 1).
It is then judged whether or not the input frame F2 is the start frame FS (step 2). When it is judged that the input frame F2 is the start frame FS, a coefficient (an initial value) representing non-conversion expressed by the following equation (5) is set as the accumulated geometric conversion coefficient MS2 between the start frame FS and the input frame F2 (step 3).
The input frame F2 is set as the preceding input frame F1, and MS2 is set as the geometric conversion coefficient MS1 between the start frame FS and the preceding input frame F1 (step 4). Further, MS2 is set as a direct geometric conversion coefficient M′S2 between the start frame FS and the input frame F2 (step 5). Thereafter, the procedure is returned to the step 1. In the step 1, the succeeding frame is read.
When it is judged in the foregoing step 2 that the input frame F2 is not the start frame FS, the geometric conversion coefficient M12 between the preceding frame F1 and the current frame F2 is calculated (step 6). The accumulated geometric conversion coefficient MS2 between the start frame FS and the current frame F2 is then calculated from MS1 and M12, as expressed by the following equation (6) (step 7).
MS2=M12MS1 (6)
The input frame F2 is set as the preceding input frame F1, and MS2 is set as the geometric conversion coefficient MS1 between the start frame FS and the preceding input frame F1 (step 8).
The direct geometric conversion coefficient M′S2 between the start frame FS and the input frame F2 is then calculated on the basis of the accumulated geometric conversion coefficient MS2 calculated in the foregoing step 7 (step 9). The details of the processing will be described later. Thereafter, the procedure is returned to the step 1. In the step 1, the succeeding frame is read. The foregoing processing is repeatedly performed up to the final frame.
First, a plurality of feature points P2 are extracted from the input frame F2 (step 11). The feature point is a point at which a pixel value between adjacent pixels is rapidly changed. Processing in the steps 12 to 14 is performed for each of the feature points.
The one feature point P2 is read (step 12). The feature point P2 is coordinate-converted into a point on the start frame FS on the basis of the accumulated geometric conversion coefficient MS2, and the point is taken as P′2 (step 13). The point P′2 is taken as an initial value for retrieval (a retrieval start point), to track a feature point on the start frame FS corresponding to the point P2, and an obtained point is taken as P″2 (step 14).
Here, the feature point is tracked using optical flow estimation by a gradient method. The optical flow estimation is based on the premise of the movement of an object in a local area. When there is a great camera movement between the start frame FS and the input frame F2, therefore, the motion of the feature point cannot be accurately detected. P′2 obtained by coordinate-converting the feature point P2 on the input frame F2 using the accumulated geometric conversion coefficient MS2, as described above, is taken as an initial value for tracking on the start frame FS. Even if there is a great camera movement between the start frame FS and the input frame F2, therefore, it is assumed that the movement is a very small movement, thereby making it possible to accurately detect the motion of the feature point.
The processing in the foregoing steps 12 to 14 is performed with respect to all the feature points extracted in the step 11, thereby obtaining the results of correspondence of all the feature points between the frames. As described above, two equations are respectively derived from correspondence points, to calculate a geometric conversion coefficient with six unknowns from all the equations using a linear solution (step 15). The calculated geometric conversion coefficient is taken as M′S2.
First, a plurality of feature points PS are extracted from the start frame FS (step 21). The feature point is a point at which a pixel value between adjacent pixels is rapidly changed. The accumulated geometric conversion coefficient MS2 is then subjected to inverse conversion (inverse matrixing), to find a coefficient MS2-inv (step 22). Processing in the steps 23 to 25 is performed for each of the feature points.
The one feature point PS is read (step 3). The feature point PS is coordinate-converted into a point on the input frame F2 on the basis of the coefficient MS2-inv, to take the point as P′S (step 24). The point P′S is taken as an initial value for retrieval (a retrieval start point), to track a feature point on the input frame F2 corresponding to the point PS. An obtained point is taken as P″S (step 25).
Here, the feature point is tracked using optical flow estimation by a gradient method. The optical flow estimation is based on the premise of the movement of an object in a local area. When there is a great camera movement between the start frame FS and the input frame F2, therefore, the motion of the feature point cannot be accurately detected. P′S obtained by coordinate-converting the feature point PS on the start frame FS using the coefficient MS2-inv obtained by subjecting the accumulated geometric conversion coefficient MS2 to inverse conversion, as described above, is taken as an initial value for tracking on the input frame F2. Even if there is a great camera movement between the start frame FS and the input frame F2, therefore, it is assumed that the movement is a very small movement, thereby making it possible to accurately detect the motion of the feature point.
The processing in the foregoing steps 23 to 25 is performed with respect to all the feature points extracted in the step 21, thereby obtaining the results of correspondence of all the feature points between the frames. As described above, two equations are respectively derived from correspondence points, to calculate a geometric conversion coefficient with six unknowns from all the equations using a linear solution (step 26). The calculated geometric conversion coefficient is taken as M′S2.
The direct geometric conversion coefficient M′S2 between the start frame FS and each of the input frames F2, which is calculated by the above-mentioned geometric conversion coefficient calculation processing shown in
Each of the frames from the start frame to the end frame is subjected to the following processing: That is, an output frame memory is first cleared (step 31). Thereafter, an object frame Fi is read in the memory (step 32). A geometric conversion coefficient M′Si corresponding to the object frame Fi is then read (step 33). A frame F′i obtained by subjecting the object frame Fi to geometric conversion using the geometric conversion coefficient M′Si corresponding thereto is then produced (step 34), and a frame F′i after the geometric conversion is rendered in the output frame memory (step 35).
It is then judged whether or not the flag for judging whether or not interpolation is required flag is set (flag=1) (step 36). When the flag for judging whether or not interpolation is required flag is not set (flag=0), interpolation processing for each pixel within a non-rendering area of the output frame memory (steps 37 to 41) is not performed. Therefore, each of the frames from the start frame and the end frame is subjected to the processing in the foregoing steps 31 to 35.
When the flag for judging whether or not interpolation is required flag is set, the interpolation processing is performed for each pixel within the non-rendering area of the output frame memory (steps 37 to 41). Description is made of the interpolation processing for a pixel at a point P in the non-rendering area.
10 frames preceding and succeeding the object frame Fi are extracted as vicinity frames Fi-neighbors. When the number of vicinity frames preceding and succeeding the object frame Fi is no more than 10, only the restricted vicinity frames are extracted. The following processing is performed for each of the vicinity frames Fi-neighbors.
That is, a geometric conversion coefficient M′Si-neighbors corresponding to the vicinity frame Fi-neighbors is read onto the memory (step 37). A coefficient M′si-neighbors-inv obtained by subjecting M′Si-neighbors to inverse conversion (inverse matrixing) is then found (step 38). Here, M′Si-neighbors-inv corresponds to a geometric conversion coefficient for coordinate-converting a point on the start frame into a point on the vicinity frame.
A point obtained by subjecting the point P to geometric conversion using M′Si-neighbors-inv is taken as Pi-neighbors (step 39) and a pixel value corresponding to the point Pi-neighbors on the vicinity frame is acquired (step 40). The acquired pixel value is taken as VPi-neighbors. Each of the vicinity frames Fi-neighbors is subjected to such processing (steps 37 to 40).
Pixel values VPi-neighbors corresponding to the points P respectively obtained from the vicinity frames Fi-neighbors are mixed, to produce the pixel value corresponding to the point P (step 41).
When the point Pi-neighbors obtained by subjecting the point P to geometric conversion using M′Si-neighbors-inv does not exist on the vicinity frames, the pixel values in the vicinity frames are not mixed. When there are no vicinity frames usable for interpolation, the interpolation is performed using a pixel value in the start frame.
The ratio of mixture of the pixel values VPi-neighbors corresponding to the points P respectively obtained from the vicinity frames Fi-neighbors is determined in the following manner.
First, the vicinity frames are previously given coefficients K such as ½n in the order of increasing distance from the object frame, as shown in
The coefficients given to vicinity frames usable for interpolation at the point P in the non-rendering area are totalized (Ktotal). An inverse number (1/Ktotal) of the total is calculated, and 1/Ktotal is added to the coefficient K given to each of the vicinity frames. This is a mixture ratio for each of the vicinity frames.
When the vicinity frames usable for interpolation at the point P in the non-rendering area are one of the vicinity frames given K=⅛and one of the vicinity frames given K=½, the total Ktotal of the coefficients given to the vicinity frames is ⅝. The inverse number of the total is 8/5. The mixture ratio for the vicinity frames given to K=⅛is (⅛)×( 8/5)=⅕, and the mixture ratio for the vicinity frames given to K=½is (½)×( 8/5)=⅘.
The mixture ratios for the vicinity frames usable for interpolation are thus found, so that the closer in time the vicinity frame is to the object frame, the higher the mixture ratio is for the vicinity frame, and the sum of the mixture ratios for the vicinity frames is one. The interpolation processing at the point P is performed with respect to all the points in the non-rendering area, to complete an output frame.
According to the above-mentioned embodiment, even if calculation of the geometric conversion coefficient ends in failure in the intermediate frame, the failure does not affect correction processing of the succeeding frames, thereby allowing high-precision automatic stabilization control to be performed. Further, the output frame after the automatic stabilization control does not include data other than image pick-up data, thereby making it possible to ensure a field of view which is not less than a field of view ensured by the input frame.
When the necessity of interpolating the non-rendering area is eliminated, the output frame memory may be cleared before the step 1 shown in
[3] Description of Modified Example of Output Frame Production Processing
Description is made of a modified example of output frame production processing.
The direct geometric conversion coefficient M′S2 between the start frame FS and each of the input frames F2, which is calculated by the above-mentioned geometric conversion coefficient calculation processing shown in
Each of the frames from the start frame to the end frame is subjected to the following processing: That is, it is first judged whether or not the flag for judging whether or not interpolation is required flag is set (flag=1) (step 41). When the flag for judging whether or not interpolation is required flag is not set (flag=0), the output frame memory is cleared (step 42). Thereafter, the program proceeds to the step 43. When the flag for judging whether or not interpolation is required flag is set (flag=1), the program proceeds to the step 43 without clearing the output frame memory.
In the step 43, an object frame Fi is read in the memory. A geometric conversion coefficient M′Si corresponding to the object frame Fi is then read (step 44). A frame F′i obtained by subjecting the object frame Fi to geometric conversion using the geometric conversion coefficient M′Si corresponding thereto is then produced (step 45), and the frame F′i after the geometric conversion is rendered in the output frame memory (step 46). The foregoing processing in the steps 41 to 46 is repeatedly performed up to the final frame.
In the output frame production processing, when the flag for judging whether or not interpolation is required flag is set, the output frame memory is not cleared for each of input frames. In the output frame memory, therefore, the results of automatic stabilization control of each of the input frames are overwritten on the results of automatic stabilization control of the past frame. The start frame is not subjected to geometric conversion. After the start frame is rendered in the output frame memory, therefore, an area corresponding to at least the size of the start frame is an area where rendering has already been performed on the output frame memory.
When the results of automatic stabilization control of each of the input frames are rendered in the output frame memory, an area other than an area where the results of automatic stabilization control are rendered enters a state where the area is interpolated by the results of automatic stabilization control of the past frame. According to this method, in the output frame, the same field of view (angle of field) as that ensured by the input frame can be ensured. When there is no moving object at the periphery of the input frame, a joint between the results of automatic stabilization control of the past frame and the results of automatic stabilization control of the current frame is not relatively conspicuous, thereby obtaining the results of automatic stabilization control having no incompatible feeling.
Although the present invention has been described and illustrated in detail, it is clearly understood that the same is by way of illustration and example only and is not to be taken by way of limitation, the spirit and scope of the present invention being limited only by the terms of the appended claims.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2003-328013 | Sep 2003 | JP | national |
| Number | Name | Date | Kind |
|---|---|---|---|
| 5210559 | Ohki | May 1993 | A |
| 6456731 | Chiba et al. | Sep 2002 | B1 |
| 6628711 | Mathew et al. | Sep 2003 | B1 |
| 6654049 | Fletcher et al. | Nov 2003 | B2 |
| 20020036692 | Okada | Mar 2002 | A1 |
| 20020131498 | Sun et al. | Sep 2002 | A1 |
| 20040005084 | Kondo et al. | Jan 2004 | A1 |
| 20050057660 | Nonaka et al. | Mar 2005 | A1 |
| 20060140507 | Ohki | Jun 2006 | A1 |
| Number | Date | Country |
|---|---|---|
| 3-3571 | Jan 1991 | JP |
| 2892685 | Feb 1999 | JP |
| 11-339021 | Dec 1999 | JP |
| 2002-330443 | Nov 2002 | JP |
| Number | Date | Country | |
|---|---|---|---|
| 20050093987 A1 | May 2005 | US |
