Obstacle Detetion In Automobile Auxiliary Safety Driving In Video Sequences

Obstacle detection is an important part of auxiliary safety system in a vehicle.Because of high speed and complex scene, the video contains a large amount of information, which leads to the difficulty to meet the requirements of real-time and accuracy at the same time, the hardware acceleration strategy increases the cost of equipment. Therefore, how to finish the target segmentation from the feasible domain in complex road environment, obtain the distinguishing characteristics, achieve real-time detection results is a difficult problem. In order to solve the problems in cost and precision, a series of theories are researched, for example, feasible region segmentation algorithm, objectness recommendation, multiscale feature pyramid, these theories are combined to get high precision and real-time obstacle detection.Firstly, on the basis of scholors’ relevant scientific research, we applies the objectness recommendation theory to detect the obstacle. Then the multiscale feature pyramid theory are used to simplify the process of feature extraction. Finally, the new algorithm for vehicle driving assist in obstacle detection and recognition are proposed.The main research work and contributions are as follows:(1) Research on the theory of image preprocessing and determine the feasible region,the relevant methods of image processing are studied, including image filtering and image enhancement. At the same time, the way to determinine the feasible region and develop a window search strategy is researched.(2) Research on the objectness recommendation theory and the SVM theory.Furthermore, the BING feature objectness evaluation method is researched, an objectness evaluation model is trained using the SVM, and then use the model to traversal windows in multiscale images. The method can significantly reduce the identification numbers. At last, we do some simulation experiments.(3) Research on the ACF features and multiscale features pyramid theory. Combined with the objectness recommendation method, a new obstacle recognition method is put forward. At last, we unfold an experimental simulation and data analysis.(4) Research on the calibration method of obstacle, the obstacle type calibration.The obstacle distance scale and position measuring method in road recognition is researched. After then, optimized method is applied to track obstacle. At last,multithread technology is used to fuse the obstacle identification and tracking. Finally,the experimental simulation data is analyzed and compared with others’ method.Experiments on the VOC data set and the actual test data show that the method proposed in this paper is feasible.