Research On Dynamic Target Identification Method Based On Binocular Stereo Vision

Binocular vision has been more and more widely used in real life.It has important implications for artificial intelligence,automated production,and medical fields.The main research content of this paper is a dynamic target identification system based on binocular stereo vision,including the identification of target shape and the extraction of depth information.This system is divided into four parts,the camera calibration in stereo vision,the tracking of the target target one by one,the stereo matching of the target image pair and the calculation of depth information.In the part of the camera calibration,the first step is build an experimental platform.The camera was calibrated by the checkerboard.The purpose of the camera calibration is to obtain internal and external parameters.In the part of tracking target,two different shapes(cylinders and cuboids)of targets were selected for the experiment.The tracking module mainly solves two problems,first tracking the position of the target one by one,and then recognising the shape of the target.In the process of tracking target one by one,how to make only one target in the image to avoid the problem that other shapes of the target influence shape recognition.The solution is to first screen the shot target image,select the area where the conveyor belt is located,and then perform preprocessing such as graying and binarization on a frame-by-frame basis.Then,the target connection area is used to track the position of the target and take it in the connected area.The largest connected area is to ensure that only one item is tracked at one frame,in other words,only one target per image frame is recognised.The main problem involved in recognising the shape of a target is when judged the shape of the target is most closest to its own shape.The solution is seeking the frame that directly under the camera in the experiment.The length and width of the target are redefined during the conveying of the target from the far to the near.The aspect ratio of the target in the frame of the images is changing.The change rule of the aspect ratio during the transmission process is studied and the transmission is found.During the process,the frame immediately below the camera is the most accurate one for recognising the shape of the target.Then,the Hough transform is used to recognising whether the shape of the object is a circle.If it is a circle,find the center of the circle and the radius in the image.In the stereo matching part,the shape of the target in the experiment is relatively simple and the features are clear.The feature matching method is used to match the image pairs.Using the FAST corner extraction algorithm to process the image,it can be seen that the extraction effect is not good,and the accumulation of the matching corners is prone to occur.In order to solve this problem,this paper improves the FAST corner extraction algorithm,which increases from 16-point detection to 24-point detection,and increases the 8 corner points around the measured pixel points.The test results are better,and the FAST corner points are extracted.The problem of corner stacking has been improved,and the target point's parallax has been obtained.In the depth information extraction module,using the calibrated internal and external camera parameters and the obtained parallax,the depth information of the target is finally obtained according to the principle of triangle distance measurement.