Research On Automotive Binocular Vision Distance Measurement System Based On FPGA

Nowadays,Binocular Vision is a more mainstream research direction in the field of automobile autonomous driving.The main research difficulties of binocular vision focus on real-time performance and the accuracy of stereo matching algorithm.A binocular vision system based on FPGA(Field Programmable Gate Array)has been designed,and the main contributions of this paper is as follows:Firstly,a binocular image acquisition system based on FPGA is designed to collect the binocular image data of the target object in the environment.The Cyclone Ⅳ series chip of Altera Corporation is selected as the development platform.Verilog hardware description language is used to write the binocular image acquisition module,temporary storage module,TFT image display module and UART network sending module,and the simulation verification of each functional module is completed.Finally,the function of binocular image acquisition and network transmission of FPGA is realized,and the acquisition rate of the system can reach 60Hz at the image resolution of 640×480.The image preprocessing algorithm based on FPGA is designed and implemented,and the median filter algorithm,Gaussian filter algorithm and Sobel edge detection algorithm are implemented based on FPGA,which improves the real-time performance of the algorithm on the FPGA platform.The delay of the whole image preprocessing system is less than 25ms,and it has high real-time performance.The semi-global matching algorithm based on dynamic programming is studied.In this algorithm,the fusion of BT generation value and X direction Sobel generation value is used as the matching cost.The cost aggregation step is completed through dynamic path planning,and the parallax calculation is realized using WTA method.The parallax optimization methods such as confidence detection,subpixel interpolation and left-right consistency detection are used to optimize the parallax graph.Finally,the results show that the semi-global matching algorithm based on path aggregation has higher matching effect and accuracy than the traditional local matching algorithm,and the stereo matching process is faster.At the end of this paper,the ranging experiment of the binocular vision ranging system has been carried out.The results show that the accuracy of the binocular ranging system is higher when the target distance is relatively close,and the mean value of the relative error can be less than 1%.In order to solve the problem that the ranging accuracy decreases when the distance is long,this paper extends the baseline of the binocular camera to 150mm,and conducts experiments after the camera calibration and image correction again.The relative error of the binocular ranging results within 10m is less than 10%,which can meet the actual needs.Compared with the traditional computer platform,the processing speed of the binocular vision ranging system is also greatly improved,and the processing speed can reach 102.5ms.