Target Recognition And Tracking Based On The Fusion Of Millimeter Wave Radar And Monocular Camera

With the development of advanced assistant driving system(ADAS)and unmanned vehicle system,the requirement of target recognition and tracking based on vehicle sensors is increasingly urgent.Although millimeter wave radar can effectively detect the position and velocity information of the target,there are still many problems such as false alarms.With the development of machine learning in recent years,Machine vision can recognize Lane lines,signal lights.Machine vision also has some problems such as poor timeliness and vulnerability to weather,night and other factors.The fusion of machine vision and millimeter wave radar can complement each other.Here we proposed a structured road model,which includes lane lines and vehicle targets.The machine vision part of the whole system includes lane recognition and tracking module,vehicle target recognition and tracking module based on CNN;A target recognition and tracking part based on millimeter wave radar,and a fusion recognition and tracking part.The main research work of this paper is as follows:(1)A lane line detection algorithm based on semantic segmentation neural network combined with clustering is proposed.The algorithm uses an end-to-end convolution neural network framework to realize the lane line image into different categories and to fit the lane line target from different types of lane line pixels.Experiments show that the algorithm can achieve high detection effect without prior lane information,and has better adaptability.(2)Aiming at the difficulty of vehicle target detection,the problem of data correlation,a vehicle target association algorithm based on Lane line.Using the neural network to realize the camera's target recognition and tracking,the effective association of different sensor targets in complex environment is realized by using the detected lane line data,and the characteristics of vehicle target and false alarm target in lane line model are used.(3)The spatial fusion model of the sensor is established by using the principle of camera imaging and the detection principle of radar,and the time fusion model of the sensor is established by interpolation extrapolation method,on the basis of which the test platform is built,and the experimental dataset is established by using the UAV for testing.Results The vehicle target fusion detection and recognition algorithm is verified in this paper.