Research On Self-localization Algorithm Of Automatic Crash Notification System For GPS Blind Area

With the continuous development of traffic,travel is more and more convenient.But at the same time,frequent traffic accidents have caused serious results.Automatic Crash Notification System(ACNS)is widely used to identify automobile accidents,collect location information,integrate and send them to rescue agencies.However,the blind area and the large error of the positioning system may greatly affect rescue efficiency.Therefore,this paper focuses on the problem of blind area and large error in positioning system.The visual system is developed to make up for the defects of the blind area,reduce the error and improve the accuracy.The software platform and hardware platform of monocular vision system were designed.The two softwares,that is,Microsoft Visual Studio 2015 and Opencv2.4.13,were combined as software platform.Cameras and lenses were selected as hardware platforms.The camera was calibrated by Zhang Zhengyou calibration method.The Local Binary Pattern(LBP)image retrieval algorithm was studied.Because of LBP noise factors may make the central pixel points change,different structural patterns may produce the same LBP coding,and cannot be adapted to the multi-scale image problem effectively.A multi-scale central LBP was proposed in this paper.Replace the neighborhood mean with the central pixel.LBP is extended to different radius neighborhoods to collect points on two radii.Then,the Corel1 k database was used for verification and analysis.Experiments show that multi-scale central LBP can effectively improve the accuracy of image retrieval.The Speed-Up Robot Features(SURF)and Random Sample Senses(RANSAC)were studied.The combination of SURF and RANSAC can delete both wrong points and correct points.SURF improved algorithm was proposed in this paper.BRIEF was used to build descriptors.Hamming distance was used to judge and carry out bidirectional matching.Grid-based Motion Statistics(GMS)was used to delete the wrong points.And then,the three methods of SIFT+RANSAC,SURF+RANSAC,SURF improvement algorithm were compared by Oxford database.Experiments’ results show that the SURF improved algorithm takes less time and retains the correct points.The self-positioning experiment of vehicle based on monocular vision was designed.A section of the campus was selected to collect images to form an image database.The multi-scale central LBP is used for image screening and the SURF improvement algorithm is used for image matching.The experimental results show that the error between the obtained results and the actual results is small and meets the experimental requirements.Based on this,the ACNS system flow was designed.