Research On Obstacle Avoidance Of Sea Unmanned Boat Based On Improved SSD

With the continuous development of unmanned technology,unmanned boat technology has been widely used in the detection of offshore islands and reefs,maritime transportation and military exploration.But in the process of the task,the ship and the iceberg floating on the sea will have a great impact on the safety of the navigation and the efficiency of the implementation of the unmanned boat because of the irregular movement and the uncertainty of the shape.Therefore,in order to ensure the unmanned boat to complete the task safely,the identification of dynamic obstacles at sea and the detection of motion path is an essential link.Due to the development of marine traffic,the improvement of weather forecast and marine map,the unmanned boat will often meet obstacles of other ships when making route to avoid obstacles such as iceberg.Therefore,the identification and tracking of the ships that affect the navigation of the unmanned boat is an important guarantee for the safe navigation of the unmanned boat.The traditional ship target detection algorithm can not classify the obstacles,and it is difficult to accurately locate the obstacles,which affects the accuracy of obstacle avoidance,so it is difficult to apply to the identification and tracking of dynamic ship obstacles.At present,the convolution neural network,which can extract the depth information and classify the objects by convolution,solves the shortcomings of the traditional algorithm,and lays a solid foundation for the recognition of the dynamic ship obstacles on the sea.Aiming at the problems of inaccurate recognition of static ship image obtained by vision system when SSD is sailing on the sea and low recognition accuracy,an improved method based on SSD(single shot multibox detector)is proposed.In this method,L2 regularization is added to the seventh layer of SSD network to improve it,which makes the parameters of each convolution layer of SSD more balanced,reduces the sensitivity of the model to local features,and improves the confidence of ship identification.In the experiment,the performance of SSD and improved SSD is compared by the accuracy and confidence.The experimental results show that this method has a better effect on marine ship identification,and the model trained based on improved SSD has a better generalization ability.Then dynamic video flow detection and path tracking are introduced into the algorithm.In the process of obstacle detection,the real-time video flow is obtained through the camera first,then the video flow is transferred from the camera to the algorithm detection module,then the detection module identifies the ship in each frame of the video flow,and then transmits each frame after identification to the tracking module for tracking point drawing After that,all the frames with recognition results and tracking points are combined to output the video stream with tracking path.Finally,the human-computer interface and the corresponding software of different platforms of the obstacle avoidance system of the sea unmanned boat based on the improved SSD are completed by using Python language.When making human-machine interaction interface,the Python language is written and run on computer and TX2 respectively by using Pycharm.Then,the PyQt5 framework is used to create the interface recognition interface of ship obstacles with various recognition functions.After the interface is generated,the Pyinstaller pair is called in Pycharm.Python programs with software interfaces are packaged to generate executable programs that are independent of the tensorflow framework.Through software package,the software is popularized,which can be used by anyone in any computer.The software which integrates the static and dynamic algorithm of ship obstacle detection constitutes the ship identification and tracking system of the sea unmanned boat.The experimental results show that the frame rate of the software can reach 30 when it is used for online real-time testing,and it can accurately identify obstacles.After the offline video stream test,the test results can be saved to the local folder at the frame rate of 25,and the test results can accurately identify and extract the obstacle driving path.These two data can meet the requirements of obstacle avoidance and identification when the unmanned boat is sailing on the sea,and can provide safety guarantee for the navigation of the unmanned boat on the sea.