Research On Intelligent Vehicle Obstacle Detection System Based On Binocular Vision

With the popularization of automobile and the rapid development of automotive electronic technology,intelligent vehicle has become an important development direction in the field of automobile,which is of great significance to improve driving safety and comfort.The environment-sensing technology is an important part of the intelligent driving technology system,which is the basis and premise of the movement obstacle avoidance,the path planning,the behavior decision-making and the motion control of the automobile.Among the many environment-sensing technologies,the vehicle binocular system has become the most important sensor in the process of environment sensing because of its advantages such as large range of information collection,high precision,strong real-time performance and so on.In vehicle binocular system,it is necessary to have both "binocular camera model" and"obstacle image".Therefore,the establishment of accurate binocular camera model is of great significance to improve the accuracy of obstacle detection.Firstly,the monocular camera mathematical model was established according to the ideal small hole imaging model.On the basis of this,the camera model was perfected by establishing the distortion model of the camera,and the key internal and external parameters of monocular camera were summarized.Secondly,based on monocular camera model and trigonometric ranging principle,the binocular ranging model and calibration model of binocular camera were established.Before obstacle detection,this paper determined that the vehicle binocular system should have the advantages of strong robustness,high accuracy and strong real-time performance,according to the actual requirements of intelligent vehicles.At the same time,this paper systematically analyzed the image preprocessing,stereo matching constraint,stereo matching algorithm system and local stereo matching flow,which were related to obstacle detection process.It laid a foundation for further research on obstacle detection algorithm.Binocular stereo matching is the main work of the disparity map of the computing environment and the obstacle detection.In order to meet the requirements of the robustness,accuracy and real-time performance of the intelligent vehicle,a local stereo matching method of the"Multi-level 3bit-Census cost calculation based on confidence interval&Adaptive cross window cost aggregation based on edge truncation&Skip typedisparity calculation based on variable maximum disparity range" frame is proposed in this paper.Experiments proved that the matching method based on the framework has stronger robustness and higher accuracy compared with the traditional matching method,and it is more suitable for vehicle-mounted binocular systems.In order to ensure the real-time and practical of the algorithm,the above algorithm was verified on the field-programmable gate array(FPGA)hardware platform.At the same time,this paper analyzed the vehicle binocular system from different angles,such as the whole structure of binocular system,data operation across clock domain,key image processing module and so on.The experiment proved that the FPGA hardware system with this algorithm can realize the image processing speed of 60fps and has high detection precision.It is proved that the algorithm and its hardware platform can be applied to intelligent vehicles.Which laid a solid foundation for the follow-up system to study the intelligent vehicle environment perception technology.