Research On Target Matching By Double Epipolar Line And Multiple Positioning Technology Based On Heterogeneous Binocular Vision

In recent years,with the continuous development of intelligent robot technology industry,artificial intelligence,image recognition,multi-source sensor fusion and other technologies are gradually applied to all walks of life.With the systematization of livestock breeding industry,inspection robot has been widely used in livestock health detection application scenarios.Patrol inspection robot can be equipped with sensing elements such as infrared thermal and high-definition color camera,so as to collect multi-source information and provide data support for livestock target health behavior analysis.Aiming at the requirements of target matching and positioning technology of heterogeneous binocular vision in livestock health inspection task,this paper innovatively proposes a detection and sorting algorithm of the same calibration point of heterogeneous images to complete the calibration of heterogeneous binocular stereo vision system;A double epipolar constrained matching algorithm in heterologous images is proposed to match the same targets in heterologous images;Combined with RFID technology and facility environment information,this paper research on the target multiple positioning technology based on heterogeneous binocular stereo vision to perceive livestock target identity information.Firstly,a stereo calibration method for HBSV system composed of infrared thermal imager and HD color camera is proposed.At the same time,this paper completes the calibration of the camera system on the inspection robot,and obtains the internal and external parameters of the HBSV system.Secondly,combined with the mathematical model of HBSV system and the characteristics of epipolar lines,a ROI region matching method for heterogeneous images is proposed;A strategy of fusing the interval constraint method and adaptive parallax constraint method is proposed to improve the target matching accuracy of heterogeneous images,it is verified that the matching effect can meet the requirements of real-time and accuracy of livestock inspection.Then,this paper uses heterogeneous binocular vision system to complete target visual localization based on 3D reconstruction technology.Combined with the structural parameter model of inspection robot,the integrated coordinate transformation mathematical model between camera,bracket,lifting mechanism,robot chassis and world coordinate system is constructed.While obtaining the pose of the target in the world coordinate system,combined with RFID positioning technology and facility breeding environment,this paper research on the multiple positioning technology to perceive livestock target identity information.Finally,the algorithm proposed in this paper is deployed to Jetson Xavier NX and deployed inside the inspection robot.The accuracy of the algorithm is verified by pedestrian target experiment.The stability and feasibility of the system were verified by field experiments in dairy and pig farms.This paper develops a visual interface based on QT,which is convenient for livestock and poultry breeding experts to access videos and analyze livestock target health.