Point Cloud Feature Extraction And Obstacle Detection In Aquaculture Waters Based On 3d Lidar

With the rapid development of Chinese aquaculture industry,aquaculture intelligence and standardization had gradually become the main development direction.Among them,the use of autonomous sailing unmanned ships for bait feeding and water quality monitoring was a low labor intensity and high degree of automation.High-precision environmental maps and obstacles in waters such as irregular water banks and aerators were the key factors affecting the autonomous operation of unmanned ships.Therefore,to obtain high-precision environmental maps and accurate obstacle information for autonomous navigation of unmanned ships had great significance.In response to the above problems,this paper carried out a study on extraction of point cloud features and obstacle detection in aquaculture waters based on 3D lidar.The main research contents include the following parts:(1)Design and construction of unmanned ships perception system: It was equipped with hardware systems on the "Salmon Three Generation" unmanned ship.Including the design of the environment-aware subsystem based on lidar,the design of its own state-aware subsystem based on GNSS / INS integrated navigation system and the design of auxiliary system.(2)Acquisition and processing of point cloud data: Used point cloud data threshold segmentation and voxel grid method to reduce the number of points and increase the processing speed of subsequent algorithms;used statistical filtering method to remove noises such as spray and bubbles generated when aerators and other equipment worked in the aquaculture waters.Linear interpolation of the combined navigation system information reduced the distortion of point cloud data.The point cloud data processing algorithm reduces the number of redundant points,removes noise such as spray and bubbles,and reduces the error of point cloud data caused by motion.(3)Real-time obstacle detection: The typical waterfront line and obstacles in the aquaculture water area were used as the detection objects,and the irregular waterfront line was fitted with Bezier curve;the box model was used to describe the obstacles in the aquaculture water area,and the obstacles were obtained by the principal component analysis The minimum bounding box model of objects,describing the size and pose information of obstacles.The test results show that the real-time obstacle detection algorithm can more accurately detect and locate obstacles in aquaculture waters,and has a better fitting effect on irregular waterfronts.(4)Point cloud map construction: In view of the error of the unmanned ship’s position and attitude information obtained by the integrated navigation system,which led to the problem of large errors in the construction of the point cloud map,the integrated navigation and the nearest iterative point algorithm were used to construct the aquaculture water map construction method.Used the latitude,longitude and attitude angle information to rotate and translate the point cloud data to the global coordinate system to complete the preliminary point cloud data matching.Considering the measurement error in the integrated navigation system,the matching algorithm was used to match the front and back point cloud data again,and the point cloud data was rotated and translated to the global coordinate system.The test results show the point cloud map obtained by the global map construction algorithm is consistent with the actual scene,and can meet the needs of the unmanned ship for the global map.(5)Design and verification of experiments: In order to verify the construction of aquaculture water area maps and obstacle detection performance,sensor accuracy test experiments and system performance measurements were carried out.Among them,the longitude and latitude error of the GNSS / INS integrated navigation system is 0.15 m,the attitude error is 0.5 °,the velocity error is 0.5m/s,and the acceleration error is 0.05m/s~2;obstacle length measurement errors of lidar under static state,yaw motion,roll motion and pitch motion are divided into 0.007 m,0.022 m,0.036 m,0.029m;map The average error of the projection experiment is 0.323m;in the waterfront detection and positioning experiment,the accuracy rate of fishpond waterfront detection and positioning is 85.97%,and the lake waterfront detection and positioning accuracy is 74.6%;obstacles in the water area In the detection experiment,the length errors measured by the simulated unmanned ship under yaw,roll,and pitch movements were 0.071 m,0.056 m,and-0.001 m,respectively,and the height errors were 0.032 m,0.014 m,and 0.030 m,respectively.The test results show that map construction and obstacle detection can basically meet the requirements of path planning and obstacle avoidance for autonomous navigation of unmanned ships.