Patent Application
20100321229
Embodiments of the present invention pertain to radar systems. Some embodiments relate to non-linear convolutional networks, and some embodiments relate to neural networks. Some embodiments of the present invention also relate to feature extraction and biometrics.
Many conventional radar systems have difficulty distinguishing slowly moving and non-moving objects from clutter. Conventional radar systems also have difficulty distinguishing persons from clutter and stationary objects even when the person is moving. Conventional radar systems also have difficulty identifying slowly moving and non-moving objects such as persons behind walls and other solid objects. These limitations of many conventional radar systems make it easy for a slow moving person to penetrate surveillance radar.
Thus, there are general needs for radar systems that can distinguish persons from non-persons. There are also general needs for radar systems that can determine whether a person is walking, sitting, standing, sleeping or performing some other action. There are also general needs for radar systems that can identify particular individuals.
A biometric radar (100) includes phase adjusters (108) to phase adjust a sequence of radar return signals received through two or more receive antennas (102) to force coherency against stationary objects. The phase-adjusting removes at least some phase noise due to the stationary objects. The biometric radar also includes a signal processor (110) to segment the phase adjusted radar return signals into a plurality of multi-resolutional Doppler components. Each multi-resolutional Doppler component may be associated with one of a plurality of biometric features (122). The biometric radar system also includes a neural network (112) to combine and weight the segmented radar returns for each biometric feature to generate weighted classifications (115) for a feature extraction process. Biometric radar system (100) may identify a person from a non-person, and in some embodiments, may further determine what the person is doing, such as whether a person is walking, talking, standing, sitting, or sleeping.
The following description and the drawings illustrate specific embodiments of the invention sufficiently to enable those skilled in the art to practice them. Other embodiments may incorporate structural, logical, electrical, process, and other changes. Examples merely typify possible variations. Individual components and functions are optional unless explicitly required, and the sequence of operations may vary. Portions and features of some embodiments may be included in or substituted for those of others. Embodiments of the invention set forth in the claims encompass all available equivalents of those claims. Embodiments of the invention may be referred to, individually or collectively, herein by the term “invention” merely for convenience and without intending to limit the scope of this application to any single invention or inventive concept if more than one is in fact disclosed.
Biometric radar system 100 may also include frequency synthesizer 116 for generating transmit waveforms for transmission by antenna assembly 102. Biometric radar system may also include timing generator 118 which generates timing signals and clock signals for the various elements of system 100.
In accordance with some embodiments of the present invention, signal processing circuitry 130 may include phase adjustors 108, signal processor 110 and neural network 112. Neural network 112 may include output layer 114, which may be a feature extractor discussed in more detail below. Output layer 114 is optional. In these embodiments, phase adjustors Phase adjusters 108 may phase adjust a sequence of radar return signals received through two or more receive antennas of antenna structure 102 to force coherency against stationary objects. The phase-adjusting may remove at least some phase noise due to the stationary objects (e.g., like walls). Signal processor 110 may segment the phase adjusted radar return signals into a plurality of multi-resolutional Doppler components. Each multi-resolutional Doppler component may be associated with one of a plurality of biometric features 122. Non-linear convolutional neural network 112 may combine and weight the segmented radar returns for each biometric feature to generate weighted classifications 115 for the feature extraction process.
In some embodiments, radar system 100 may identify a person from a non-person, and in some embodiments, radar system may further determine what the person is doing. For example, radar system 100 may determine whether a person is walking, talking, standing, sitting, or sleeping i.e., the person's positional state. In some embodiments, radar system 100 may determine whether the identified person is carrying an object and/or whether there are multiple persons. In some embodiments, radar system 100 may be able to identify specific individuals. These embodiments are discussed in more detail below.
In some embodiments, radar system 100 may perform a range scan of the environment and may lock one radar range gate on static objects, such a wall. The other radar range gates may be place on the target. In these embodiments, the radar range gate locked on the static objects may be used as a phase reference to force coherency for the system. In some embodiments, movement of the operator may also be removed allowing hand-held operation.
An example of phase adjusting and removal of phase noise to help force coherency suitable for use with some embodiments of the present invention is disclosed in U.S. patent application No. 10/[TBD] (Docket No. PD-05WO01) filed [DATE], which incorporated herein by reference, although the scope of the invention is not limited in this respect. In some embodiments, a neural network configured to operate similarly to those disclosed by Yann LeCun et al in the article entitled “Gradient-Based Learning Applied to Document Recognition” Proc. of the IEEE, November 1998, may be suitable for use as neural network 112, although the scope of the invention is not limited in this respect.
In some embodiments, each multi-resolutional Doppler component may have a frequency structure 120 associated with one of biometric features 122. Biometric features 122 may be associated with humans and comprise heartbeat, breathing, postural sway and gait, although the scope of the invention is not limited in this respect.
In some embodiments, the neural network 112 processes the segmented and phase-adjusted radar return signals to identify a person from a non-person, although the scope of the invention is not limited in this respect. In some embodiments, weighted classifications 115 correspond to one or more positional states of a person including a person walking, a person talking, a person standing, a person sitting, a person sleeping, multiple persons, and a person with an object, although the scope of the invention is not limited in this respect. In these embodiments, radar system 100 may identify not only a person from non-human objects, but may identify the person's positional state.
In some embodiments, signal processor 110 further segments the phase-adjusted radar return signals into non-biometric features including static objects (e.g., walls) and clutter, although the scope of the invention is not limited in this respect. In some embodiments, signal processor 110 may perform a discrete Fourier transform DFT, such as a Fast Fourier transform FFT, as part of the segmentation to measure and identify predetermined frequency components of a particular biometric feature. In some of these embodiments, the segmentation may include using a matched filter to segment known responses e.g., particular frequency characteristics.
In some embodiments, signal processing circuitry 130 may also include phase processors to lock a radar range gate on static objects and place other radar range gates on target objects (e.g., a human) to help force coherency. The phase processors may be part of the phase adjusting operations performed by phase adjustors 108.
In some embodiments, further comprising output layer 114 may comprise a feature extractor and may perform a feature extraction process to generate output 115 selected from one of the positional states.
In some embodiments, signal processor 110 integrates the radar return signals over differing coherency times for the different of the multi-resolutional Doppler components of the plurality. In these embodiments, some of the multi-resolutional Doppler components have longer coherency times than other multi-resolutional Doppler components. For example, signal processor 110 may integrate the multi-resolutional Doppler components associated with the biometric features of heartbeat and breathing for longer coherency times and may integrate the multi-resolutional Doppler components associated with the biometric features of postural sway and gait for shorter coherency times, although the scope of the invention is not limited in this respect. In some embodiments, signal processor 110 may integrate the radar returns for coherency times as great as 20-40 seconds, for example, for detecting heartbeat and detecting breathing, while integration may be performed on the radar returns for as little as ⅓ of a second, for example, for detecting postural sway and gait. In some embodiments, the segmenting performed by signal processor 110 may generate structured packets for each multi-resolution Doppler component.
In some embodiments, non-linear convolutional neural network 112 is trained with a back-propagation learning process to set weights of the biometric features 122 associated with the positional states. The learning process is described in more detail below. In some embodiments, neural network 112 may comprise other types of artificial neural networks such as a support vector machine, although the scope of the invention is not limited in this respect. In these embodiments, the support vector machine may classify with as much margin as possible and may be more robust to changes in the environment providing less false classifications.
In some embodiments, neural network 112 comprises a combination of a plurality of non-linear convolutional networks. Each of the non-linear convolutional networks may be associated with one of weighted classifications 115 and may correspond to one of the positional states, although the scope of the invention is not limited in this respect. In these embodiments, signal processor 110 may direct the multi-resolution Doppler components (e.g., based on it's frequency structure) associated with particular biometric features 122 to one or more of the convolutional networks of the plurality based on the classifications 115. The combination of the convolutional networks may performs a multi-layer (i.e., multi-level) non-linear convolutional process on the directed multi-resolutional Doppler components using non-linear filters to generate weighted positional states for a human target. Examples of this process are described in more detail below.
In some embodiments, different patterns of segmented data may be processed and directed to the different convolutional networks. For example, a person sitting may have a heartbeat and breathing, while a person standing may also have postural sway, while a person walking would have a faster heartbeat, a certain velocity and a gait. For example, the multi-resolutional Doppler component associated with heartbeat and breathing may be directed to a convolutional network associated with a person sitting, the multi-resolutional Doppler components associated with heartbeat, breathing and postural sway may be directed to a convolutional network associated with a person standing, and the multi-resolutional Doppler components associated with heartbeat, breathing, gait and slow velocity may be directed to a convolutional network associated with a person walking.
In some embodiments, the non-linear convolutional networks of the plurality are configured to operate in parallel and are weighted to the quantity and quality of information in the multi-resolutional Doppler components (e.g., structured packets).
In some embodiments, neural network 112 may be further trained with the back-propagation learning process to set weights of biometric features 122 associated with a particular individual. Accordingly, particular individuals may be identified by system 100 based on their individual biometric features, such as their unique body features including their individual gait, postural sway, heartbeat and breathing, the combination being unique to an individual, although the scope of the invention is not limited in this respect. In these embodiments, one of the convolutional networks of the plurality may be associated that particular individual. In this way, radar system 100 may be trained to recognize persons from non-persons, as well as trained to recognize a person's positional state. In some embodiments, radar system 100 may be trained to further recognize an increased heart rate and associate the increased heart rate with an object (e.g., a weapon) to identify a threat. For example, radar system 100 may be training to recognize a person wearing an improvised explosive device vest (IED Vest) in front of or behind walls by the radar signature cause by the metal around the person's chest area. In another example, radar system 100 may be trained to recognize persons carrying rifles behind a wall from the unique radar signature resulting from the swaying guns.
In some embodiments, antenna structure 102 may have a transmit antenna and two or more receive antennas. The two or more receive antennas may receive the sequence of radar return signals in at least both azimuth and elevation, although the scope of the invention is not limited in this respect. Radar transmitter 100 may also include frequency synthesizer 116 to generate waveforms for transmission by the transmit antennas. The transmitted signals may include a series of short pulses stepped in frequency and corresponding to the received sequence of radar return signals. In some embodiments, the short pulses may have a frequency ranging between 3 and 6 GHz, a bandwidth of approximately 125 MHz, and a length of about 8 nanoseconds (nsec), although the scope of the invention is not limited in this respect. In some embodiments, the radar system disclosed in U.S. patent application No. 10/[TBD] (Docket No. PD-03W185) and incorporated herein by reference may be suitable for portions of radar system 100, although the scope of the invention is not limited in this respect. In some embodiments, radar system 100 may be able to see a target, such as a person) through or around objects based on the selection of the frequency range and the selection of the characteristics of the series of short pulses that are stepped in frequency. In these embodiments, a person and/or a person's positional state may be identified behind opaque objects and within buildings.
Although radar system 100 is illustrated as having several separate functional elements, one or more of the functional elements may be combined and may be implemented by combinations of software-configured elements, such as processing elements including digital signal processors (DSPs), and/or other hardware elements. For example, some elements, such as signal processing circuitry 130, may comprise one or more microprocessors, DSPs, application specific integrated circuits (ASICs), and combinations of various hardware and logic circuitry for performing at least the functions described herein. In some embodiments, the functional elements of radar system 100 may refer to one or more processes operating on one or more processing elements. In some embodiments, frequency structures 110 and classifications 115 may be stored in a system memory.
In accordance with some embodiments, the segmented data may be directed to one or more non-linear convolutional networks. Furthermore, more than one non-linear convolutional network may be used for different detection tasks. In these embodiments, the processes performed in parallel by the non-linear convolutional networks may automatically determine the best features to discriminate biometrics based on the quantity and quality of the returns. In some embodiments, a-prior information is used to determine the combinations of the convolutional networks for particular biometric features.
In some embodiments, neural network 112 (
Once the training data is segmented, the topology may be designed by combining multiple structures into the multi-layer convolutional networks. For example, breathing and heartbeat components may be combined in a network for a person sitting. The network design may be completed by adding non-linear filters with weights, which may be selected through the learning or training process, illustrated in
Unless specifically stated otherwise, terms such as processing, computing, calculating, determining, displaying, or the like, may refer to an action and/or process of one or more processing or computing systems or similar devices that may manipulate and transform data represented as physical (e.g., electronic) quantities within a processing system's registers and memory into other data similarly represented as physical quantities within the processing system's registers or memories, or other such information storage, transmission or display devices.
Embodiments of the invention may be implemented in one or a combination of hardware, firmware and software. Embodiments of the invention may also be implemented as instructions stored on a machine-readable medium, which may be read and executed by at least one processor to perform the operations described herein. A machine-readable medium may include any mechanism for storing or transmitting information in a form readable by a machine (e.g., a computer). For example, a machine-readable medium may include read-only memory (ROM), random-access memory (RAM), magnetic disk storage media, optical storage media, flash-memory devices, electrical, optical, acoustical or other form of propagated signals (e.g., carrier waves, infrared signals, digital signals, etc.), and others.
The Abstract is provided to comply with 37 C.F.R. Section 1.72(b) requiring an abstract that will allow the reader to ascertain the nature and gist of the technical disclosure. It is submitted with the understanding that it will not be used to limit or interpret the scope or meaning of the claims.
In the foregoing detailed description, various features are occasionally grouped together in a single embodiment for the purpose of streamlining the disclosure. This method of disclosure is not to be interpreted as reflecting an intention that the claimed embodiments of the subject matter require more features than are expressly recited in each claim. Rather, as the following claims reflect, invention may lie in less than all features of a single disclosed embodiment. Thus the following claims are hereby incorporated into the detailed description, with each claim standing on its own as a separate preferred embodiment.
