Research On Vehicle Obstacle Avoidance Method Based On Binocular Vision

Vehicle obstacle avoidance system based on binocular vision consists of three main parts: perception,decision-making and control.Among them,the purpose of perception is to accurately detect and segment the obstacles in front of the vehicle,and to measure the speed and distance;the purpose of the decision-making is to determine the driving direction according to the detection results of the target lane obstacles by the active binocular platform.The purpose of control is to distribute the results of the decision to the underlying actuator,accurately control the vehicle to go straight or change lanes,and avoid obstacles.Based on previous studies,this paper focuses on obstacle detection and segmentation and vehicle obstacle avoidance model,and realizes large-scale real-time detection and accurate segmentation of generalized obstacles.Combined with binocular rotation platform and vehicle obstacle avoidance model,the multi-path driving of vehicles is realized,the main research work of this paper is as follows.Firstly,the visual sensor is used as the sensing element of obstacle detection.The imaging model and projection principle of the camera are studied,and the selection and installation position of the camera are introduced.The binocular camera is calibrated and corrected by Zhang Zhengyou calibration method to obtain the internal and external parameters and distortion parameters of the camera.Secondly,in the aspect of obstacle detection,VIDAR(Vision-IMU based detection and range method)is used to identify the height of the feature points of the obstacles in the images extracted based on MSER(Maximally Stable Extremal Regions)algorithm,and the pseudo obstacle feature points are eliminated.Then,according to the regional location of different obstacles,the obstacles are segmented based on morphological closed operation,K clustering and monocular and binocular vision.Using the characteristics of the same obstacle and the same feature point,the FFT template matching,DBSCAN clustering and threshold determination method are used to locate,track and match the template points,calculate the height and verify the threshold results.Experiments show that the obstacle detection and segmentation method based on VIDAR and binocular vision can provide a larger detection field of view and improve the accuracy and robustness of obstacle detection and segmentation.Thirdly,in terms of vehicle obstacle avoidance,the active binocular platform and vehicle obstacle avoidance model are designed.When the distance between the vehicle and the obstacle in front meets the lane change distance,the obstacle detection of the target lane is realized by the rotation of the binocular camera.According to the detection results,the vehicle is controlled to change lanes or go straight.When the distance between the vehicle and the obstacle in front meets the emergency braking distance,the vehicle performs emergency braking to ensure the safety of the vehicle.Aiming at the straight driving condition of the vehicle,the typical scene of emergency braking is simulated and verified by constructing the Car Sim/Simulink joint simulation platform.The path planning based on A* algorithm and the tracking control based on MPC(Model Predictive Control)algorithm are verified by theory and simulation.The simulation test verifies the feasibility and effectiveness of the vehicle obstacle avoidance model designed in this paper,which can meet the safe driving of the vehicle.Finally,a real vehicle test platform is built to verify the obstacle detection and segmentation method proposed in this paper.On the basis of obstacle detection,according to the vehicle obstacle avoidance model,the vehicle straight driving condition and the vehicle lane changing condition are designed and verified respectively.The integrated navigation system is used to track the driving trajectory.The real vehicle test data show that the obstacle avoidance success rate is 87.5% and 88.2% respectively when the vehicle is in emergency braking mode and lane changing mode.The experimental results show that the proposed method meets the driving requirements in different scenarios.