| In this paper,the obstacle detection and recognition method in vehicle environment perception system is taken as the research object.Since the information acquisition of single sensor must have certain deficiencies,this paper uses the technical route of machine vision and millimeter wave radar information fusion to realize the detection and identification of obstacles in front of the vehicle.The main work is as follows.1)Based on analyzing the measurement environment of millimeter wave radar,the nearest target of the same lane is selected as radar primary target.Aiming at the unknown or time-varying statistical characteristics of system noise in target tracking,combined with the traditional Sage-Husa adaptive filtering algorithm and square root filtering algorithm,an improved linear adaptive square root Kalman filter algorithm is proposed to achieve target state estimation and simulation.The results verify the accuracy and stability of the algorithm.Based on the consistency test to effective target,the life cycle algorithm is used for decision-making of effective target.2)Aimed to make full use of the context information of the network,the deconvolution module is introduced in the network model of the traditional SSD algorithm.The experiment verifies effectiveness of the improved algorithm.Compared with other experimental algorithms,the improved algorithm has higher precision and better real-time performance.3)The conversion relationship is established between the millimeter wave radar coordinate system,the camera coordinate system,the world coordinate system,the image coordinate system,and the image pixel coordinate system.Technique for Camera Calibration propsed by Zhang Zhengyou is used to achieve the internal and external parameters and distortion parameters of the camera,while the distortion correction formula is used to correct the nonlinear distortion of the ideal linear model.On the basis of the fusion of radar and camera information,the target detected by the millimeter wave radar is projected onto the synchronous image and a dynamic region of interest is established.The effectiveness of the multi-sensor fusion technology proposed in this paper is verified by comparing the obstacle detection and recognition algorithm with or without radar assistance. |