Research On Vision Algorithm Of Intelligent Cleaning Robot For Water Environment

The current situation of plastic pollution in the water environment is extremely severe,which has seriously threatened the survival of aquatic organisms and human health.In order to solve the increasingly serious plastic pollution problem in the water environment,vision-based intelligent cleaning robots have been designed in our lab.For robots,the visual system is essential,Among the vision system,the target detection algorithm plays a vital role.The main purpose of this thesis is to propose two target detection algorithms for the water surface environment and the underwater environment.Experimental results show that the detection speed and accuracy of the improved algorithm are better than other target detection algorithms.For the purpose of improving the intelligent robot vision system,this thesis adds binocular image matching,multi-target tracking,and binocular measurement after the target detection algorithm.The multi-modules work together to efficiently realize the functions of the intelligent robot vision system.The field and pool experiments show that the results obtained by the vision system can provide real-time and reliable target information for the robot,and help the robot to complete the garbage collection autonomously in the complex water environment.This thesis has important theoretical research significance and application value in the field of computer vision and the research of intelligent robot.This thesis studies the related issues involved in the vision system of the intelligent cleaning robot for water environment.The main research contents are as follows:1.Through a large number of investigations and analysis on the current mainstream traditional detection algorithms and machine learning-based target detection algorithms,this paper uses the improved YOLOv3 algorithm to achieve water surface garbage detection based on the characteristics of the water surface environment,application scenarios,and the characteristics of the robot hardware design.In virtue of improve the performance of real-time detection,this paper transforms the detection of YOLOv3 from three-scale to two-scale.In addition,in order to ensure the accuracy of detection,this paper also re-clustered the bounding boxes of the water surface garbage data set to replace and update a part of the original YOLOv3 anchor boxes that are not applicable to this data set.With the help of the proposed detection method,the water surface garbage detection algorithm has high-speed and high-precision target detection capabilities,and the experimental results show that the achieved target detection algorithm meets the expected requirements.2.Aiming at underwater garbage detection,this thesis proposes a garbage target detection method based on improved YOLOv4,which can achieve high-speed and high-precision target detection.More specifically,the YOLOv4 algorithm is selected as the basic neural network for underwater garbage detection.In order to further improve the detection speed,channel pruning and layer pruning are performed on the trained YOLOv4 model.The fine-tuning mechanism helps the pruned model to restore the accuracy.Experimental results show that even if the pruned YOLOv4 parameters are only 7.06%of the original model,the detection can still maintain high performance.With the help of improved detection methods,the robot has the ability to collect garbage autonomously.3.For the purpose of consummating the functions of the intelligent robot vision system,this article adds binocular image matching,multi-target tracking,and binocular measurement after target detection.The four parts are reasonably and efficiently integrated into a complete vision system.Field experiments and pool experiments have successfully verified the efficiency of the vision system.