Research On Collision Avoidance Technology For Safty Autonomous Driving

With the development of artificial intelligence,intelligent changes are taking place in all walks of life.Such as autonomous driving,more and more scholars are engaged in autonomous driving.Collision avoidance is the crucial part of autonomous,thus this paper is focus on the collision avoidance technology.The past anti-collision detection system usually focuses on the abnormal detection of moving objects.According to the research,it is found that not only moving objects but also sometimes stationary objects may pose a threat to the vehicle.To solve this problem,this paper presents an anti-collision detection method based on vehicle video.In this paper,anti-collision detection is consist of object detection and object dangerous classification.On the one hand,the calculation of the initial motion model of moving object is optimized by semantics segmentation.On the other hand,based on the initial motion model of moving object and image super-pixel information,the energy equation of scene flow based on time consistency of current frame is constructed,and the label and operation of moving object in current frame can be obtained.Then,we fuse the object regions based on motion information,depth information and semantic information.Finally,according to the dangerous areas in front of the vehicle and the driving areas in front of the vehicle,the objects in the scene can be classified.The main work of this dissertation is as follows:1)A method of scene flow calculation based on time consistency is presented.Firstly,the initial object area of the scene is calculated by in-depth learning;Secondly,the center of the initial object area is used as the clustering center of the three-dimensional feature point,and the three-dimensional feature point is clustered to calculate the initial motion model of the moving object in the current frame;Thirdly,the motion label and the motion model of the moving object in the previous frame are input to calculate the current frame of the moving object in the current frame.Finally,the energy equation of scene flow based on time consistency is constructed according to the information of the previous calculation.The energy equation is minimized by conditional random field,and the scene flow of the current frame is obtained,and the label of the moving object in the current frame and the corresponding motion model are obtained.2)A multi-information object detection and dangerous status classification method is presented.Firstly,the stereo object region is obtained by calculating the UV-disparity,and obtained the object region with semantic information.Secondly,according to the self-motion model of the vehicle,the dangerous area in front of the vehicle is calculated,and the objects in the dangerous area are obtained.Finally,the vehicle driving area is calculated and the objects in the dangerous area are classified.3)At last,we design a collision avoidance system based on the above work.We firstly introduce the overall design of the system.And we explain the implementation of system related functions.Finally,we analyze and evaluate the effectiveness of the system.