Development And Research Of Binocular Vision Distance Measurement System Based On Embedded AM5728

Distance measurement based on a dual camera system originated from bionics that human perceive depth is through a process called binocular vision stereopsis or stereo vision.The dual cameras capture the same scene,obtain reconstruction 3D information and then according to the binocular system triangulation relationship to calculate the distance of the target to the camera.Compared with the conventional distance measurement method,the dual camera system has advantages of simple equipment,interference from the transmission medium,which can be widely used in robot vision navigation,visual detection and positioning,the industrial robot hand-eye system and other important areas.However,the method of the dual camera system has the problems of high complexity,large computational complexity and precision and so on,so it difficult to miniaturization and used in the field of industrial production and robot navigation.In order to solve the existing problem of real-time processing speed is slow for a general processor,and binocular image registration algorithm is complicated,a dual cameras system of distance measurement system based on an embedded AM5728 is developed and researched in this thesis.The specific work includes:(1)In the hardware development aspect,the AM5728-based binocular vision ranging hardware platform was designed and developed for the first time,and the algorithm was studied on this platform.AM5728 on embedded processor as the core processing unit of image processing,combined with power management,binocular USB camera,HDMI display auxiliary module and PC communication,etc.(2)In the algorithm aspect,the relative theoretical basis of camera distortion correction,binocular calibration,image matching,image calibration,binocular matching and ranging calculation is studied in the developed hardware system.Then an acceleration strategy of the adaptive intra-frame and inter-frame image matching is designed.Consequently,the search speed of image matching and the match effect are improved,and the accuracy of measurement distance between the camera to the targets is improved through optimizing the kalman filter algorithm.The model of binocular vision range system is constructed.(3)In the software aspect,the design of the human-computer interaction interface is completed through QTcreate and OpenCV joint programming under Linux system,and the integration and embedded tasks of the algorithm on the hardware platform based on AM5728 are completed through the GCC cross-compiler to compile system kernel and file system.(4)The prototype is designed through testing the hardware and the related software algorithms,and the rationality of the design of hardware and software are verified.The experiment results for distance measurement using the binocular vision system show that the system can realize the simultaneous acquisition of binocular images,video source switch,the depth of the image display and the output of distance measuring results for the image with the resolution of 640 × 480,and the processing speed of the system can be Up to 25 fps.Analysis of the Experimental results show that the optimal measurement range of the system is from 0.2 to 1.1 meter and the relative error of distance measurement is less than 5 %.It can meet the general application requirements.