Search Robot Design And Environment Perception Method Based On Multi Robots Visual Observation

This Dissertation studies the special environment search problem,which mainly solves the two directions of narrow environment and large target environment search.The narrow environment search is solved by designing a structure in line with the environmental characteristics,and the environment search with large targets is solved by multi robots visual observation and perception of the environment.The main research works in this Dissertation is as follows:(1)The structure of robot mobile platform suitable for unstructured environment is analyzed and compared,and a multi legged retractable mobile robot platform structure imitating caterpillar is designed.For the core structure of the robot-body turning structure and walking foot,the overall structure of the robot is designed from the aspects of principle design,structure design,motion state analysis and force analysis.(2)Based on the research of previous scholars,a method of dynamic camera height calibration based on the combination of image imaging principle and coordinate system transformation is proposed.This Dissertation studies the basic principle,specific implementation steps,parameter calculation method,calibration experiment,calibration result inspection and analysis,and compares this method with the traditional calibration method in the actual calibration experiment.(3)Using oriented FAST and rotated BRIEF(ORB)feature points as matching and fusion features,the information transmission framework between multiple robots is established,the camera relative pose calculation method based on simultaneous observation and the simultaneous localization and mapping(SLAM)method after RGB image fusion of multiple robots are established,and the experiment after video image fusion is carried out.In the experiment,it is found that when the number of sampling frames is at least 15 frames/s,The simultaneous localization and mapping point cloud can be generated.(4)After optimizing the key frame selection and loop closure of the traditional ORB-SLAM algorithm,a dynamic key frame selection method based on proportional differential(PD)controller and an ORB feature matching method based on affine transformation are proposed,a point cloud image fusion process based on scene recognition is established,and a consistency analysis and evaluation method for the fused point cloud map is proposed.(5)The coupling of you only look once(YOLO)algorithm and ORB-SLAM algorithm is used for semantic annotation,and the locally convex connected patches(LCCP)algorithm is used to segment the point cloud to establish a three-dimensional point cloud semantic map.Taking two robots creating semantic point cloud map based on ORB-SLAM as an example,the algorithm process is given,and the feasibility of semantic map creation process is verified by experiments.There are 83 figures,10 tables and 106 references in this Dissertation.