Vehicular vision system

[0001] This application is a continuation of U.S. patent application Ser. No. 10/461,699 filed Jun. 13, 2003, now U.S. Pat. No. 7,263,209 the entire disclosure of which is incorporated herein by reference.

[0002] 1. Field of the Invention

[0003] The present invention relates to vehicular vision systems. In particular, the present invention relates to a method and apparatus for providing target detection to facilitate collision avoidance and/or mitigation.

[0004] 2. Description of the Related Art

[0005] Collision avoidance systems utilize a sensor for detecting objects in front of an automobile or other form of vehicle. The sensor may be a radar, an infrared sensor or an optical vision system. The sensor generates a rudimentary image of the scene in front of the vehicle and, by processing that imagery, obstacles can be detected within the imagery.

[0006] These collision avoidance systems identify that an obstacle exists in front of a vehicle, but do not classify the obstacle nor provide any information regarding the movement of the obstacle. As such, the driver of the vehicle may be warned of an obstacle or the automobile may take evasive action regarding an obstacle, yet that obstacle may present no danger to the vehicle.

[0007] Therefore, there is a need in the art for a method and apparatus that provides a vehicular vision system that classifies obstacles to facilitate obstacle avoidance.

[0008] The invention provides a vision system for a vehicle that identifies and classifies objects (targets) located proximate a vehicle. The system comprises a sensor array that produces imagery that is processed to generate depth maps (or depth images) of the scene proximate a vehicle. The depth maps are processed and compared to pre-rendered templates of target objects that could appear proximate the vehicle. A target list is produced by matching the pre-rendered templates to the depth map imagery. The system processes the target list to produce target size and classification estimates. The target is then tracked as it moves near a vehicle and the target position, classification and velocity are determined. This information can be used in a number of ways. For example, the target information may be displayed to the driver, or the information may be used for an obstacle avoidance system that adjusts the trajectory or other parameters of the vehicle to safely avoid the obstacle. The orientation and/or configuration of the vehicle may be adapted to mitigate damage resulting from an imminent collision, or the driver may be warned of an impending collision.

[0015] FIG. 1 depicts a schematic diagram of a vehicle 100 utilizing a vision system 102 to image a scene 104 that is located proximate vehicle 100. In the embodiment shown, the imaged scene is in front of the vehicle 100. Other applications of the system 102 may image a scene that is behind or to the side of the vehicle. The vision system 102 comprises sensor array 106 coupled to an image processor 108. The sensors within the array 106 have a field of view that images a target 110 that is located in front of the vehicle 100. The field of view of the sensors in a practical system may be ±12 meters horizontally in front of the automobile (e.g., approximately 3 traffic lanes), a ±3 meter vertical area and provides a view of approximately 40 meters in front of the vehicle.

[0016] FIG. 2 depicts a block diagram of the hardware used to implement the vision system 102. The sensor array 106 comprises, for example, a pair of optical cameras 200 and 202 and an optional secondary sensor 204. The secondary sensor 204 may be a radar transceiver, a LIDAR transceiver, an infrared range finder, sonar range finder, and the like. The cameras 200 and 202 generally operate in the visible wavelengths, but may be augmented with infrared sensors, or they may be infrared sensors themselves without operating in the visible range. The cameras have a fixed relation to one another such that they can produce a stereo image of the scene.

[0017] The image processor 108 comprises an image preprocessor 206, a central processing unit (CPU) 210, support circuits 208, and memory 212. The image preprocessor 206 generally comprises circuitry for capturing, digitizing and processing the imagery from the sensor array 106. The image preprocessor may be a single chip video processor such as the processor manufactured under the model Acadia I™ by Pyramid Vision Technologies of Princeton, N.J.

[0018] The processed images from the image preprocessor 206 are coupled to the CPU 210. The CPU 210 may comprise any one of a number of presently available high speed microcontrollers or microprocessors. The CPU 210 is supported by support circuits 208 that are generally well known in the art. These circuits include cache, power supplies, clock circuits, input-output circuitry, and the like. Memory 212 is also coupled to the CPU 210. Memory 212 stores certain software routines that are executed by the CPU 210 to facilitate operation of the invention. The memory may store certain databases 214 of information that are used by the invention as well as store the image processing software 216 that is used to process the imagery from the sensor array 106. Although the invention is described in the context of a series of method steps, the method may be performed in hardware, software, or some combination of hardware and software.

[0019] FIG. 3 is a functional block diagram of the functional modules that are used to implement the present invention. The sensors 200 and 202 provide stereo imagery to a stereo image preprocessor 300. The stereo image preprocessor is coupled to a depth map generator 302 which is coupled to the target processor 304. The target processor receives information from a template database 306 and from the optional secondary sensor 204.

[0020] The two cameras 200 and 202 are coupled to the stereo image preprocessor 300 which, for example, uses an Acadia I™ circuit. The preprocessor 300 calibrates the cameras, captures and digitizes imagery, warps the images into alignment, and performs pyramid wavelet decomposition to create multi-resolution disparity images. Each of the disparity images contains the point-wise motion from the left image to the right image. The greater the computed disparity of an imaged object, the closer the object is to the sensor array.

[0021] The depth map generator 302 processes the multi-resolution disparity images into a two-dimensional depth image. The depth image (also referred to as a depth map) contains image points or pixels in a two dimensional array, where each point represents a specific distance from the sensor array to point within the scene. The depth image is then processed by the target processor 304 wherein templates (models) of typical objects encountered by the vision system are compared to the information within the depth image. As described below, the template database 306 comprises templates of objects (e.g., automobiles) located at various positions and depth with respect to the sensor array. An exhaustive search of the template database may be performed to identify a template that most closely matches the present depth image. The secondary sensor 204 may provide additional information regarding the position of the object relative to the vehicle, velocity of the object, size or angular width of the object, etc., such that the target template search process can be limited to templates of objects at about the known position relative to the vehicle. If the secondary sensor is a radar, the sensor can, for example, provide an estimate of both object position and distance. The target processor 304 produces a target list that is then used to identify target size and classification estimates that enable target tracking and the identification of each target's position, classification and velocity within the scene. That information may then be used to avoid collisions with each target or perform pre-crash alterations to the vehicle to mitigate or eliminate damage (e.g., lower or raise the vehicle, deploy air bags, and the like).

[0022] FIG. 4 depicts a flow diagram of a method 400 showing the operation of the present invention. The method 400 begins with the setup and calibration of the cameras at step 402. At step 404, the method captures and digitizes the images from the cameras. At step 406, the imagery generated from each of the cameras is warped into alignment to facilitate producing disparity images. At step 408, the method 400 generates multi-resolution disparity images from the camera images using pyramid wavelet decomposition. Steps 402, 404, 406 and 408 are performed within an off-the-shelf stereo image preprocessing circuit such as the Acadia I™ circuit. The multi-resolution disparity image is created for each pair of frames generated by the stereo cameras. The disparity image comprises, in addition to the disparity information, an indication of which of the disparity pixels in the image are deemed valid or invalid. Certain disparity values may be deemed invalid because of image contrast anomalies, lighting anomalies and other factors. This valid/invalid distinction is used in processing the depth image as described below.

[0023] At step 410, the multi-resolution disparity image is used to produce a depth map. This transformation is not required but in the present embodiment it simplifies subsequent computation. The depth map (also known as a depth image or range image) comprises a two-dimensional array of pixels, where each pixel represents the depth within the image at that pixel to a point in the scene. As such, pixels belonging to objects in the image will have a depth to the object and all other pixels will have a depth to the horizon or to the roadway in front of the vehicle.

[0024] To confirm that an object exists in the field of view of the cameras, step 412 may be implemented to utilize a secondary sensor signal for target cueing. For example, if the secondary sensor is a radar, the sensor produces an estimate of the range and position of the object. As such, the template matching process will require less time since the template search will be restricted to the radar provided position and depth estimate.

[0025] Steps 414, 416, 418, 420 and 422 are used to search a template database to match templates to the depth map. The database comprises a plurality of pre-rendered templates, e.g., depth models of various types of vehicles that are typically seen by the vehicle. In one embodiment, the database is populated with multiple automobile depth models at positions in a 0.25 meter resolution 3-D volume within the scene in front of the vehicle. In this embodiment, the vertical extent of the volume is limited due to the expected locations of vehicles on roadways. The depth image is a two-dimensional digital image, where each pixel expresses the depth of a visible point in the scene with respect to a known reference coordinate system. As such, the mapping between pixels and corresponding scene points is known. The method 400 employs a depth model based search, where the search is defined by a set of possible vehicle location pose pairs. For each such pair, a depth model of the operative vehicle type (e.g., sedan or truck) is rendered and compared with the observed scene range image via a similarity metric. The process creates an image with dimensionality equal to that of the search space, where each axis represents a vehicle model parameter, and each pixel value expresses a relative measure of the likelihood that a vehicle exists in the scene with the specific parameters.

[0026] Generally, an exhaustive search is performed where the template is accessed in step 414, then the template is matched to the depth map at 416. At step 418, a match score is computed and assigned to its corresponding pixel within the image where the value (score) is indicative of the probability that a match has occurred. Regions of high density (peaks) in the scores image indicate the presence of structure in the scene that is similar in shape to the employed model. These regions (modes) are detected with a mean shift algorithm of appropriate scale. Each pixel is shifted to the centroid of its local neighborhood. This process is iterated until convergence for each pixel. All pixels converging to the same point are presumed to belong to the same mode, and modes that satisfy a minimum score and region of support criteria are then used to initialize the vehicle detection hypotheses. At step 420, the target list is updated if the match scores are large enough to indicate that the target has been identified.

[0027] The match score can be derived in a number of ways. In one embodiment, the depth difference at each pixel between the template and the depth image are summed across the entire image and normalized by the total number of pixels in the template. In another embodiment, the comparison (difference) at each pixel can be used to determine a yes or no “vote” for that pixel (e.g., vote yes if the depth difference is less than one meter, otherwise vote no). The yes votes can be summed and normalized by the total number of pixels in the template to form a match score for the image. In another embodiment, the top and bottom halves of the template are compared separately to the depth image. At each pixel, if the value of the template depth is within one meter of the value of the depth image, a yes “vote” is declared. The votes in the top and bottom image halves are summed separately to provide a percentage of yes votes to the total number of pixels. The top and bottom percentages are multiplied together to give a final match score.

[0028] At step 422, the method 400 queries whether another template should be used. If another template should be used or the exhaustive search has not been completed, the method 400 returns to step 414 to select another template for matching to the depth map. The templates are iteratively matched to the depth map in this manner in an effort to identify the object or objects within the scene.

[0029] In one embodiment, during the template matching process, the process speed can be increased by skipping ahead in larger increments of distance than typically used depending upon how poor the match score is. As such, normal distance increments are ¼ of a meter but if the match score is so low for a particular template than the distance may be skipped in a larger increment, for example, one meter. Thus, a modified exhaustive search may be utilized. When the exhaustive search is complete, method 400 continues to optional step 424. The secondary sensor information is used to confirm that an object does exist. As such, once the target is identified, the secondary sensor information may be compared to the identified target to validate that the target is truly in the scene. Such validation reduces the possibility of a false positive occurring. At step 424, the target list from the vision system is compared against a target list developed by the secondary sensor. Any target that is not on both lists will be deemed a non-valid target and removed from the target lists.

[0030] At step 426, the target size and classification is estimated by processing the depth image to identify the edges of the target. The original images from the cameras may also be used to identify the boundaries of objects within the image. The size (height and width) of the target are used to classify the target as a sedan, SUV, truck, etc. At step 428, the target and its characteristics (boundaries) are tracked across frames from the sensors. A recursive filter such as a Kalman filter may be used to process the characteristics of the targets to track the targets from frame to frame. Such tracking enables updating of the classification of the target using multiple frames of information.

[0031] At step 430, the method 400 outputs target position, classification and velocity. This information can be used for pre-crash analysis by a vehicle's collision avoidance system to enable the vehicle to make adjustments to the parameters of the vehicle to mitigate or eliminate damage. Such processing may allow the automobile's attitude or orientation to be adjusted, (e.g., lower or raise the bumper position to optimally impact the target) the air-bags may be deployed in a particular manner to safeguard the vehicle's occupants with regard to the classification and velocity of target involved in the collision, and the like.

[0032] While the foregoing has described a system that uses a multi-resolution disparity image (or map) to produce a depth map in step 414, as previously noted this is not required. For example, the dashed lines in FIG. 4 illustrate a method in which the multi-resolution disparity image produced in step 408 is used directly, rather than for the production of a depth map, step 410. As illustrated, after generation of the multi-resolution disparity image in step 408, a secondary sensor is used for target cueing, step 412, to confirm that an object exists in the field of view of the cameras.

[0033] Then, the template database, which now comprises a plurality of pre-rendered templates of multi-resolution disparity images, e.g., disparity images of various types of vehicles, is searched, step 414. A match test is then performed to match the templates to the multi-resolution disparity image, step 415. Then, a match score is computed, step 418, and the target list is updated, step 420. A decision is then made whether another template is to be used, step 422. If so, a loop is made back to step 414.

[0034] While the foregoing is directed to embodiments of the present invention, other and further embodiments of the invention may be devised without departing from the basic scope thereof, and the scope thereof is determined by the claims that follow.