Study On Stereo Object Detection Around The Vehicle Based On Vision

Car collision accident accounted for 80%of vehicle accidents,96%of the traffic accident is caused by the consciousness of the people,the drivers lacking of attention to the surrounding environment and can't fully monitor the surrounding environment is the main cause of these collision accidents.The C-NCAP which began to implement from 2018 raise the proportion of active safety in the evaluation system up to 15%.In addition to the FCW and LDW,the AEB and pedestrian protection(daytime)were added to the evaluation system,active safety has become an important part of car safety.The Advanced Assisted Driving System(ADAS)can only detect and alert people and vehicles in front of the vehicle and in a certain area.The Around View Monitor(AVM)System only monitor the environment around the vehicle and doesn't detect the stereo objects.In the integrated system of ADAS and AVM,there is an area around the vehicle that only monitors but does not carry out object detection.In order to enhance the active perception of the surrounding objects,it is better to help the driver to observe the surrounding environment of the vehicle,In this paper,a stereo object detection method based on single camera 3D reconstruction is studied,A stereo object detection method using polar line constraint in the active safety integrated system is proposed to help reduce the occurrence of collision accident.In this paper,ADAS and AVM integrated system in the four directions of the vehicle camera,access to the image around the vehicle,through the camera calibration to obtain internal and external parameters.The feature points of the image are extracted by the SIFT corner detection algorithm.The pyramid optical flow method is used to track and match the feature points.The matching matrix is used to calculate the basic matrix.Based on the 3D reconstruction,a method of three-dimensional object detection using polar line constraint is proposed.Sum of absolute differences is used to extract the exact matching point to calculate the exact base matrix.The projection line is calculated from the base matrix.The feature points of stereo objects are classified by field segmentation method,mark the classification of the three-dimensional objects,the adjacent mark area to fuse,to avoid the same three-dimensional objects were marked many times,Finally,the stereo objects which were detected in the single camera scene projected into the panorama,marked in the panoramic display of stereo objects,A complete detection algorithm is designed for this idea of stereo object detection,and the algorithm is implemented on the S32V234 development board.The hardware is accelerated to make the algorithm reach real-time operation.