Research On Binocular Vision System Based On UAV

In recent years,with the rapid development of UAV technology.The realization of automation and intelligence of UAV,has gradually become a popular research direction.In order to realize autonomous navigation and obstacle avoidance functions of UAV in specific application fields,it is necessary to use binocular stereo vision technology to help UAV acquire distance information in the scene.Therefore,how to combine UAV technology with binocular vision technology has become an increasingly worthy research topic.Based on the characteristics of UAV,this paper designs a binocular vision system that can calculate the distance information in the application scene for UAV.It mainly carries out the following three aspects of work:(1)Research and improvement of binocular stereo matching algorithm: to realize binocular vision technology,it is necessary to obtain the distance information in the scene through binocular stereo matching algorithm.Firstly,analyze the binocular stereo matching algorithm,and summarize the whole operation process mainly includes the following steps: the initial matching cost calculation,matching cost aggregation,parallax selection and image optimization.Aiming at the problem of high complexity of image information in UAV application scene,the traditional Census transformation algorithm relies too much on the information of the window center point,which may lead to certain limitations in the process of initial matching cost calculation.Therefore,the improved Census algorithm is adopted by the author,which makes use of the correlation of pixel information of each pixel in the matching window to find the matching point more accurately,so as to achieve a better matching effect.In the MATLAB simulation tool,imports the image set of the public database Middlebury,obtains the matching result of the traditional Census algorithm and the improved Census algorithm,and tests and evaluates the two algorithms.The experimental results show that the matching precision can be improved by using the improved Census algorithm to calculate the initial matching cost.(2)Design of hardware algorithm architecture: Aiming at the problem that the traditional processor based on instruction set structure is inefficient in image processing,and it is difficult to meet the processing speed requirement of UAV binocular vision system,the Field Programmable Gate Array(FPGA)is selected for algorithm deployment.On the basis of theoretical analysis of binocular stereo matching algorithm,use High-level Synthesis(HLS)to designed the FPGA binocular stereo matching IP core.Through the analysis of the simulation results and evaluation report generated by the tool,the correctness of the function of binocular stereo matching IP core is verified,and it is concluded that the operation rate of binocular stereo matching IP core is able to meet the real-time requirements of the UAV binocular vision system.(3)Selection of hardware platform and the actual hardware test: select the FPGA hardware platform based on xc7z020clg484-1 chip to finish the actual hardware test.Import the binocular stereo matching IP core into Vivado development tool,and display the results calculated by FPGA on the screen after the test program is downloaded to FPGA.The final test results show that the UAV is able to complete the calculation flow of binocular vision system through FPGA in practical application.