Realization Of Binocular Vision Measurement Method Based On Zynq Platform

The distance measurement technology based on stereo vision is widely used in transportation,military,aerospace and other fields.Compared with traditional measuring methods,it has the advantages of non-contact,high efficiency and anti-interference.Binocular vision is an important branch of stereoscopic vision technology.Compared with monocular vision,it has a larger amount of information and can achieve 3D reconstruction.Most of the early binocular vision studies were based on the PC platform.This could not be applied to harsh and dangerous application environments,or didn't meet the requirements for flexibility and portability.In view of the above situation,binocular measuring technology based on embedded platforms has been paid attention to and developed rapidly.Compared with single ARM,DSP,FPGA and other platforms,the Zynq platform integrates the FPGA+ARM architecture on a single chip,which has the advantages of good scalability,high performance,low power consumption,and low cost.This paper studies and implements a binocular vision measuring method based on Zynq.This paper studies the binocular vision model.It expounds the principle of binocular measurement technology,then calculates three-dimensional information through stereo calibration and correction.We collect and display images through the FPGA resources of the Zynq platform,and store and deal with them effectively.This paper simulates the practical application scene.It and color space markers and threshold segmentation methods are used to identify markers with color features.The overall recognition performance is improved by filtering and morphological processing methods.This paper uses the improved single-point correction method to achieve the center pixel correction,more simply,fast,and saving resources.Then,by extracting the pixels at the center of the marker,the parallaxes generated by the different camera angles of the left and right cameras are calculated.In combination with the binocular version model,according to the triangulation principle,the three-dimensional information of the marker is recovered.For image preprocessing operations,the high-level synthesis tool provided by Xilinx is fully utilized to encapsulate it into a hardware IP core.The hardware acceleration of image preprocessing is realized,and the effect is obvious.The requirement for real-time image processing is initially achieved.Finally,the experimental verification of Zynq-based binocular measuring algorithm is carried out.In an attempt to verify the feasibility and effectiveness of the algorithm,we measure and analyze the relative movement distances in different directions.After analyzing the experimental data,we find that the relative movement of the target in the X and Y directions can be detected precisely through the proposed method.The measurement error is within 10 mm.The measurement error in the Z direction is limited by the system model and changes dynamically with depth,within a reasonable range.