Research And Development Of Wheeled Robot Based On Binocular Vision Real-time 3D Reconstruction

Mobile robots(wheel type,crawler type,foot type,etc.)can replace humans to complete dangerous tasks such as planetary exploration,undersea operations,disaster relief,and can also help people do daily chores and provide entertainment.In the context of intelligent manufacturing 2025,the intelligent wave of industrial equipment such as wheeled handling robots,parallel palletizing robots,and multi-joint welding robots has swept across various industries.Vision-based 3D reconstruction technology has been rapidly developed as an important part of the equipment’s intelligent perception of the environment.Among them,the binocular vision 3D reconstruction technology simulates the principle of human eye imaging,and uses the parallax of the left and right eyes to reconstruct the target in 3D,which can effectively improve the product detection efficiency in industrial production and reduce production costs.Based on the 3D reconstruction algorithm based on artificial features,this topic studies the real-time dense 3D reconstruction technology of binocular vision,and builds an experimental platform for the 3D reconstruction algorithm for wheeled mobile robots.Firstly,starting from the most basic principle of camera imaging,various camera imaging models and the mutual transformation between different coordinate systems are studied.The causes of camera distortion and its corresponding solutions are analyzed.Aiming at the problem that the accuracy of the binocular ranging system decreases with the increase of the object distance,a variable baseline binocular vision hardware system is built.Through the horizontal comparison of the mainstream calibration schemes,the TOOLBOX＿calib toolbox with higher accuracy is selected as the calibration method in this thesis.Secondly,the structural framework and operation principle of the mainstream SLAM(simultaneous localization and mapping)system are studied and analyzed,and the ORB-SLAM2 algorithm with high tracking accuracy and concise programming language is selected as the visual odometry part of the real-time 3D reconstruction system.Keyframe and camera information(camera＿info)output interface.The selected key frames are used as input images to reduce the computational complexity of the real-time reconstruction system.Then,the principle of stereo matching is analyzed,and the reconstruction of dense point cloud based on three stereo matching algorithms of ELAS,SGBM and BM is realized.On the basis of ensuring the quality of the point cloud and taking into account the computing time,the ELAS algorithm is selected as the stereo matching part of the reconstruction system.At the same time,combined with the camera trajectory output by visual odometry in real time,a local point cloud splicing node is written.Finally,an experimental platform for the 3D reconstruction algorithm of a wheeled robot with four-wheel independent drive is built.The wheeled robot is stably controlled by the adaptive PID closed-loop control algorithm integrated with fuzzy control,and the precision experiment of the binocular vision real-time dense 3D reconstruction algorithm is carried out on the wheeled robot experimental platform.