Patent Application
20080112595
The invention described and claimed hereinbelow is also described in German Patent Application DE 102006053286.4 filed on Nov. 13, 2006. This German Patent Application, whose subject matter is incorporated here by reference, provides the basis for a claim of priority of invention under 35 U.S.C. 119(a)-(d).
The invention relates to a method for detecting image regions that are conspicuous in terms of the movement in them in a sequence of images of a scene; an apparatus, in particular for performing this method; and a corresponding computer program.
Video surveillance systems are used for monitoring such areas as streets, railroad stations, industrial plants, buildings, or squares with video cameras, and on the basis of the sequences of images recorded with the video cameras, discovering anomalies in the monitored areas.
While in earlier times the evaluation of the sequences of images was usually done by trained monitoring personnel, this monitoring activity is increasingly being either taken over or at least supported by computers. The computers use image processing algorithms for automatic evaluation of sequences of images. A typical principle for extracting moving objects from a monitored scene is to separate moving objects from the—essentially static—background of the scene, track them over time, and in the event of relevant patterns of motion, trip alarms. For object segmentation, particularly in the context of detecting the object in a sequence of image, the differences in the image between a current camera image and a so-called scene reference image that models the static background of the scene, are typically evaluated. Such image processing algorithms are described for instance in the scientific article by K. Toyama, J. Krumm, B. Brumitt, and B. Meyers: “Wallflower: Principles and Practice of Background Maintenance” in ICCV 1999, Corfu, Greece.
From other fields in image processing, namely video data compression, it is known to use what is known as the optical flow. For example, International Patent Disclosure WO 2005/006762 A2 defines the optical flow as the distribution of motion speeds of brightness patterns in an image, and uses the optical flow to improve the compression in video data streams. However, nothing about object detection or object tracking is found in this reference.
It is therefore an object of the present invention to provide a method for detecting image regions that are conspicuous in terms of the movement in them as well as an apparatus and a computer program for performing the method, which are further improvements of the existing solutions.
The method of the invention serves to detect image regions that are conspicuous in terms of the movement in them in a sequence of images of a scene. The scene is in particular the area being recorded with a video camera or the like and can be stationary, so that the viewing angle or field of view of the camera is unchanged over time. Alternatively, the scene may also be in motion, for example being recorded by a camera that is pivoted back and forth. A sequence of images is a series of images that have been taken one after the other, and in particular each with the same time interval between them.
Based on the sequence of images of the scene, a current optical flow field is calculated that includes motion vectors of image regions of the scene. The motion vectors, in the simplest case, are embodied as location vectors, which describe the translation of the image regions from one image to the next or to a subsequent image in the scene, and/or as speed vectors, which describe the direction of the motion of the image regions as well as the speed or apparent speed of the image regions from one image to the next or over the course of the sequence of image. The optical flow field is time-dependent and is preferably—particularly in order to suppress interference—chronologically filtered. Preferably, one optical flow field is generated for each new image in the image sequence.
In a further step, which may also be performed before the determination of the current optical flow field, a main flow field is determined that includes main flow directions of image regions of the scene. This main flow field serves as a reference flow field, and for image regions of the scene it indicates main flow directions that correspond to a “normal” motion in the scene. For instance, the main flow direction of an image region of a scene in which a moving escalator is shown, is in the direction of motion of the escalator, since typically both the people and the steps of the escalator are moving in that direction of motion.
According to the invention, from all the image regions of the scene, certain image regions are selected in which the motion vectors are oriented differently from the main flow directions. The selected image regions form image regions that are conspicuous in terms of the movement in them and are used for constructing a counterflow field. The counterflow field thus contains solely motion vectors of image regions that are conspicuous in terms of the movement in them.
The invention is based on the thought that the known methods for detecting relevant motion patterns, which are based on separating the moving objects from the background of the scene, function well, as long as individual objects can be distinguished well enough from one another. However, when the number of objects is becoming higher and higher and these objects are getting closer and closer together, or in other words when the density of objects in the scene is increasing, the individual objects can no longer be distinguished from one another, and segmented regions in the image of various objects fuse into one common object. However, once error-free object segmentation is no longer possible, object tracking cannot be performed successfully any longer, either, and in the known method it finally fails.
According to the invention, object segmentation is first dispensed with, and the sequence of images is examined using the optical flow, so as to attain an analysis of the motion of small regions of the image beyond the sequence of image. In this way, image regions can be detected that for instance flow counter to a main flow direction, and thus objects can be recognized that move counter to the main flow. The proposed invention thus utilizes information that occurs from calculation of the motion of small image regions. For the calculation of these flow fields, it is not necessary to have a reference image of the scene, since only the current camera images are needed.
In a preferred embodiment of the invention, the image regions of the current flow field and/or of the optical main flow field and/or of the counterflow field are embodied as pixels, blocks, or segments. A first option is to process the various fields pixel by pixel, so that for each pixel, a motion vector or a main flow direction is for instance defined. A further option is to define blocks that in particular each have the same size, so that the scene, that is, the flow field, main flow field and/or counterflow field, is regularly subdivided. A further option is subdivision into segments that may be arbitrarily defined. For instance, the aforementioned escalator may form one segment in the counterflow field. In particular, it should be pointed out that the flow field, main flow field and counterflow field need not have the same grouping. For instance, it is within the scope of the invention that the optical flow field is calculated pixel by pixel, while conversely the optical main flow field is grouped into segments.
In one possible alternative embodiment, the main flow field is predetermined either a priori or manually by a user. This alternative embodiment is preferred if, based on the scene being observed, “normal” directions of motion include the user, as in the example of the aforementioned escalator.
Alternatively, the main flow field may also be either generated or updated by the current optical flow field. In the generation of the main flow field, it is provided in particular that the main flow field is formed by means of filtration, and in particular chronological averaging, of successive current optical flow fields. In the case of updating, the main flow field is updated by means of the current flow field, which in particular is filtered, in which already-calculated information is preserved for the most part.
Optionally, an aging algorithm may be used, in which the main flow field is fundamentally generated or updated via chronological averaging, but in addition logical conditions are checked, so that a reversal of the main flow, for instance, can be detected and a change in the main flow field can be initiated.
The counterflow field includes or comprises image regions in which the motion vectors and the main flow directions are each located differently from one another. The degree of deviation is in principle adjustable as a parameter, as a function of the application. In particular in image regions of the counterflow field, the motion vectors and the main flow directions are contrary and/or perpendicular to one another and/or are in an intermediate state between them. In alternative embodiments, it may be provided that even a slight deviation in the orientation, for example over 100, in particular over 200, and preferably over 300, suffices for a corresponding image region to be classified as being conspicuous in terms of motion and to be ascribed to the counterflow field.
In a preferred refinement of the invention, in the counterflow field among the image regions, counterflow regions that are cohesive with the same or similar direction of motion and/or speed are determined. This combination of individual image regions into one counterflow field region corresponds to segmentation in the known image processing methods. However, because the segmentation is performed in the counterflow field and not in the individual images of the sequence of image, the sources of error are minimized. Additional security is attained if it is not merely checked whether the direction of motion matches, or matches sufficiently, but also if the common speed of motion is taken into account. In this way, counterflow field regions that with high certainty cohere can be determined that represent an object, such as a person, even if that object is partially concealed by other objects.
In a refinement of the method, the cohesive counterflow field regions are tracked over time. In particular, for each newly arriving image in the sequence of image, a new current optical flow field is generated, and the cohesive counterflow field regions are detected. In this way, time-dependent trajectories of the cohesive counterflow field regions can be generated, and—as a result—an object detected in this way can be tracked.
In a preferred refinement of the method, upon the occurrence of a cohesive counterflow field region and/or in the event of predefined motion patterns of a cohesive counterflow field region, an alarm is tripped.
The invention furthermore relates to an apparatus for detecting image regions that are conspicuous in terms of the movement in them in a sequence of images of a scene, having the characteristics of claim 8, the apparatus preferably being embodied for performing the method that has just been described.
The apparatus is preferably implemented in the form of a computer or some other programmable device. In particular, the apparatus forms part of a video surveillance system, which besides the apparatus has a plurality of video cameras and/or video monitoring recorders, which are connected to the apparatus by cable or power cord and/or in wireless fashion, for instance via the internet.
The apparatus has a calculation module, which is embodied for calculating a current optical flow field in the scene; the optical flow field includes motion vectors of image regions of the scene. The motion vectors may be embodied as location vectors and/or as speed and/or acceleration vectors.
A determination module is furthermore provided, for determining the optical main flow field; the main flow field describes the main flow directions of image regions of the scene. The determination module may generate or update the optical main flow field on the basis of previous current optical flow fields of the scene; alternatively, the determination module is embodied as an interface or input device, so that a main flow field can be defined by a user.
A generation module is embodied in terms of programming and/or wiring so as to generate a counterflow field, and the counterflow field is formed from motion vectors of image regions that are conspicuous in terms of the movement in them. The motion vectors of the image regions that are conspicuous in terms of the movement in them are oriented differently from the main flow directions of the same image regions.
Optionally, the apparatus includes a segmentation module, which segments image regions having the same or similar motion vectors in the counterflow field to form a cohesive counterflow region.
A likewise optional tracking module is embodied for tracking the detected counterflow region over time, or in other words over further images in the sequence of image, and in particular for generating a trajectory of the counterflow region.
Also optionally, an identification module is provided, which makes it possible to identify the detected or generated counterflow region. In the identification, an association of the counterflow region with an object class is made, for instance persons, car, a suitcase, dog, or the like.
In an advantageous refinement of the apparatus, the apparatus has an alarm module, which is embodied for tripping an alarm upon the detection of a counterflow region and/or if the ascertained trajectory of the counterflow region matches predetermined motion patterns.
A further subject of the invention is a computer program with program code means which is embodied for performing the method on a computer system and/or on an apparatus in accordance with the present invention.
The novel features which are considered as characteristic for the present invention are set forth in particular in the appended claims. The invention itself, however, both as to its construction and its method of operation, together with additional objects and advantages thereof, will be best understood from the following description of specific embodiments when read in connection with the accompanying drawings.
a, 2b, and 2c are views showing a main flow field, a current flow field and a resultant counterflow field illustrating the method in accordance with the present invention;
a and 4b are views showing an application of the apparatus in accordance with the present invention in observing a scene in form of approach to a set of stairs;
Based on the camera 3, sequences of images are carried via the interface 2 to a calculation module 6, in which, on the basis of the sequences of images, the optical flow of the scene shown in the image regions is calculated. In the calculation of the optical flow, image regions, such as pixels, freely definable segments, or predefined blocks, are tracked from one image to another of the sequence of image. From the results of the calculations, motion vectors are ascertained, which describe the motion of the image regions from image to image, for instance in terms of image coordinates. In a simple embodiment, the motion vectors are defined as translation vectors, that is, location vectors which represent the offset of the applicable image region from one image to another.
In more-complex applications, the motion vectors are formed as speed and/or acceleration vectors. Particularly in the latter case, the optical flow may also be called an estimate of the speed vectors, projected onto the image plane, of visible objects, which are based on brightness patterns in the image. The totality of all the motion vectors of the scene shown in the sequence of images is called the optical flow field and is updated incrementally for newly arriving images in the sequence of image.
A determination module 7 determines an optical main flow field that includes main flow directions of image regions of the scene. The optical main flow field corresponds to a reference field for the ensuing determination of image regions that are conspicuous in terms of the movement in them. The main flow directions correspond to the directions of motion of objects, image regions, etc., which in the course of monitoring are classified as normal.
In a generation module 8, a counterflow field is generated, on the basis of the current optical flow field and of the determined optical main flow field.
For better comprehension of the generation of the counterflow field, see
b, in a similarly schematic view, shows a current optical flow field 11, as it has been calculated from the sequence of images of a scene. A majority of the motion vectors 12 of the optical flow field are oriented identically and correspond in direction to the motion vectors 10 of the main flow field 9. In the central region, however, a group of motion vectors 13 can be seen that are oriented contrary to the motion vectors 12. These motion vectors 13 represent the motion of objects, such as persons, who are moving counter to the main flow, or in other words against the stream. In the generation module 8 (
Since the image regions used in the various fields 9, 11 and 14 may be quite small, it is possible in this way for even partly concealed objects that are moving counter to a main flow to be detected. The principle is not to segment the object in the original image in the sequence of images but instead first to extract image regions that have patterns that are conspicuous in terms of motion and to examine or segment the extracted image regions in a further step.
There are various alternatives for determining the main flow field 9. In one, the main flow field may be input manually by a user, for instance into the database 5 (
As can be seen again most simply from
In a downstream tracking module 16, the counterflow regions formed are tracked over time as tracked objects. To that end, the apparatus 1 is operated sequentially, and for each newly furnished image of a sequence of image, one current optical flow field is generated, and—following the procedure described—a new counterflow region is formed. On the basis of the counterflow regions or tracked objects being tracked, a trajectory of the counterflow regions or tracked objects is formed that is then evaluated in an alarm module 17. The alarm module compares the trajectory or already compares the determined counterflow region with predetermined alarm rules and triggers an alarm signal in accordance with the alarm rules.
The updated main flow field 9 is subdivided—for instance by user input—into blocks or segments, and a mixed grouping in the main flow field 9 is also possible. Particularly upon grouping into segments, these may be predetermined selectively from outside—that is, by a user—or generated by means of segmentation of the main flow field, and image regions with similar main flow directions are combined. As a result, a main flow field 9 divided up into blocks or segments is output and then used for detecting the image regions that are conspicuous in terms of the movement in them.
a and 4b show the application of the apparatus 1 in observing a scene in the form of an approach to a set of stairs. The main flow direction in the scene is oriented from top to bottom in the plane of the drawing as indicated by the arrow 18. In accordance with the main flow direction, nearly all the persons 19 shown are moving in that direction. In a first step, based on a sequence of images of this scene, the optical flow field 11 of the scene is determined and after that filtered with the main flow field, which has only motion vectors or directional vectors in the direction of the arrow 18.
As the outcome of the filtering, an image region 20 is extracted that shows a person who is running up the stairs counter to the main flow direction 18. As can be seen from
In summary, the analysis of optical flow fields over time is a powerful tool for recognizing an object moving against the main stream. This recognition is possible particularly in scenes that are full of people, but naturally can also be applied in other scenes as well.
It will be understood that each of the elements described above, or two or more together, may also find a useful application in other types of constructions differing from the type described above.
While the invention has been illustrated and described as embodied in a method for detecting image regions that are conspicuous in terms of the movement in them; apparatus and computer program for performing the method, it is not intended to be limited to the details shown, since various modifications and structural changes may be made without departing in any way from the spirit of the present invention.
Without further analysis, the foregoing will so fully reveal the gist of the present invention that others can, by applying current knowledge, readily adapt it for various applications without omitting features that, from the standpoint of prior art, fairly constitute essential characteristics of the generic or specific aspects of this invention.
| Number | Date | Country | Kind |
|---|---|---|---|
| 102006053286.4 | Nov 2006 | DE | national |
