Research On Obstacle Detection And Avoidance System For Unmanned Vehicle On Water Surface

With the acceleration of globalization and the impact of the COVID-19 epidemic in recent years,the economy of marine has gradually become the main pillar of global trade,and the marine equipment technology,which plays an important role in promoting the marine economy,has gradually become intelligent.Unmanned Surface Vehicles(USV),as an important branch of marine equipment,are widely used in marine transportation and missions.The ability to perceive and avoid obstacles autonomously is an important prerequisite for safe navigation of USV.Therefore,conducting research on how to accurately perceive surface obstacles and autonomous obstacle avoidance technology is of great significance for the development of unmanned aerial vehicle technology.Combining computer vision and binocular stereo vision,this thesis studies the surface obstacle detection and abstacle avoidance system of Unmanned Surface Vehicles.The specific contents are as follows:(1)Improve the target detection algorithm to realize the detection of obstacles on the water surface.In order to improve the perception of USV of obstacles on the water surface,this thesis proposes a complex weather data enhancement method to enhance the self-made data set of obstacles on the water surface,and combines the transfer learning method to train YOLOv3 algorithm to obtain the detection algorithm of obstacles on the water surface A series of improvements and optimizations have been made to address the issues of slow convergence and insufficient accuracy.The experimental results show that the improved algorithm in this thesis improves detection accuracy by 5.19% and reduces parameter size by 44% while ensuring detection speed.It balances detection accuracy and model volume,and enhances the generalization ability of the detection algorithm for complex weather conditions.(2)According to the binocular stereo vision principle,the distance measurement of objects is realized.Accurate positioning of objects in front of USV is the key to autonomous obstacle avoidance Based on the binocular vision theory,the parameters of binocular camera are calibrated and the stereo correction BM stereo matching algorithm is constructed.According to the disparity map generated by the matching algorithm and the distance calculation formula,the depth of field estimation of the object is realized.Finally,the ranging experiments in different distance ranges are designed to verify the effectiveness of the ranging system The experimental results show that the binocular ranging system can locate objects in a certain range in three-dimensional space and the error is within 5%(3)Designs the USV software and hardware control system and obstacle avoidance strategy.Firstly,we designs and builds the hardware system of the USV water surface obstacle detection and obstacle avoidance system,and use Python to complete the control program on the main control chip.Secondly,the upper computer application software of the system was developed using the.NET Framework framework using C # language,achieving data communication and information exchange between USV and clients.Finally,an obstacle avoidance strategy based on detection and ranging was designed,and a mathematical model was established to analyze and explain the obstacle avoidance process,verifying the reliability of the obstacle avoidance strategy.(4)The experimental platform of unmanned ship is set up for functional verification experiment.The catamaran unmanned ship is selected and the experimental platform of real boat is set up to design different surface object detection and single and multiple surface object collision avoidance experiments The experimental results verify the feasibility and applicability of the surface object detection and obstacle avoidance system proposed in this thesis.