| The development of information technology in fishing ports is an important part of fishery technology innovation,and the construction of fishery information technology can improve the efficiency and economic benefits of fishery production.Usually,when fishing vessels return to port,the catch needs to be loaded,unloaded and transferred in time to ensure the freshness of the catch.In traditional fishing ports,the transfer of catch is mostly done manually,which has low efficiency and safety risks.The application of self-driving technology to the catch transfer vehicles in fishing ports can improve the safety and efficiency of catch transfer operations,and efficiently complete the task of catch transfer to maximize the freshness of the catch to enhance economic benefits.However,in order to achieve autonomous driving,it is necessary to accurately sense the environment of the fishing port first.Usually,there are limitations in the detection range and blind spots of individual sensors,and the fusion of sensory information from multiple sensors can achieve the complementary advantages and disadvantages of different sensors.Combining multiple sensors for the same sensing system can effectively improve the accuracy and safety of the environmental sensing system and better cope with various complex scenarios in the fishing port environment.Based on this background,this paper designs a fishing port environment sensing system based on the fusion of millimeter wave radar and image sensor information to achieve the detection and identification of catch transfer vehicles in fishing port environment,which provides a technical basis for the application of automatic driving technology for catch transfer vehicles.First,this paper analyzes the characteristics of millimeter wave radar data,designs a data filtering algorithm to filter out invalid data in millimeter wave radar point cloud data,and improves the clustering algorithm on this basis to achieve the acquisition of valid targets in radar detection data.Secondly,considering the performance and generalization of the image detection algorithm,this paper selects the YOLOv5 S neural network model as the basic detection framework,and improves the problem that the model is not effective in detecting small-sized targets by optimizing the structure of the multi-scale feature network,while applying the attention mechanism to the neural network model to enhance the effect of the network in extracting feature information from images.After verification,the overall performance of the improved network model has been improved accordingly.Finally,a monocular distance estimation model is applied on the basis of the image detection network to realize the longitudinal distance detection of targets based on a single image sensor.Then,this paper analyzes the hierarchical structure of sensor fusion,realizes the spatial alignment between millimeter-wave radar and image sensors based on the coordinate system conversion relationship,and designs a temporal matching algorithm to complete the temporal alignment between the two sensors,and combines the region of interest established by the radar detection results in the image plane with the image detection results by a method based on the cross-parallel ratio combined with radial distance matching to achieve data association between millimeter-wave radar and image sensors,and establishes a target detection framework based on the fusion strategy of millimeter-wave radar and image sensors.Finally,by building an experimental platform integrating millimeter-wave radar and image sensors,data under various working conditions were collected in different scenarios in the fishing port,and the detection results of individual sensors were compared with those of the fusion system to verify the effectiveness as well as the advantages of the fusion system based on millimeter-wave radar and image sensors designed in this paper for the detection method of catch transfer vehicles in the fishing port environment. |
PDF Full Text Request