Recognition And Tracking Of AUV Target Aiming System

China's sea area is vast and has very rich aquatic resources and submarine mineral resources.Therefore,it is very necessary to explore and protect the marine resources.The Autonomous Underwater Vehicle(AUV)is an important branch of underwater robots and the visual aiming and tracking system carried on it is of great significance to ocean exploration and protection.Besides,visual aiming and tracking system could be applied to both military and non-military field such as military guidance,enemy reconnaissance,fish exploration or tubing tracking,etc.The research mainly designs a target aiming system,the main contents of the research are as follows:First of all,camera calibration and image preprocessing of video acquisition are required.In order to facilitate the cooperation with control and tracking mechanism,the camera is calibrated to obtain the internal parameter matrix.As the matrix changes,the angle required by the motion mechanism change is obtained.Because the visual part of target aiming system is easily affected by environmental factors such as lighting conditions,in preparation for the detection algorithm,image filtering and edge extraction processing are performed on the captured video.The second part is about research on target detection algorithms.Based on the target characteristics of the target aiming system,this paper analyzes the detection algorithm of Hough circle detection and studies the YOLO series of target detection models,YOLOv1,YOLOv2 and YOLOv3,with real-time and accuracy into consideration.As the detection target is single and have high requirements on detection speed,based on pooling layer idea in YOLOv1,anchor idea in YOLOv2 and multi-scale prediction idea in YOLOv3,this paper finally chooses to use YOLOv3-tiny as the target detection algorithm.The training results show that the detection accuracy of YOLOv3-tiny is 91.4%.Then,the target tracking algorithm is studied.This part mainly studies the target tracking algorithm based on KCF.Since the KCF tracking algorithm cannot solve the problem of target occlusion in actual use,this paper uses a Kalman filter to predict and correct the target's moving position,which can be effectively solve the target occlusion problem.The last part is overall design and verification of target aiming system through the experiment carried out by Raspberry Pi 3B with Open CV vision library and Py Torch framework.The result shows that target aiming system has a favorable aiming accuracy for both static and moving targets.