Research On Underwater Multi-targets Location Based On Vision And Proximate Sensor

Nowadays, the underwater technology, especially for the self-operating system, the research on underwater multi-targets recognition and positioning are particularly important. In this paper, in order to boost the capabilities of recognition and positioning and enhance the operational capability and safety performance of the means of delivery, the research on multi-targets positioning based on multi-sensors such as vision sensor and sonar are presented.At first, the operating system for multi-targets positioning are established based on an open-frame micro-underwater robots.By theoretical analysis, simulation and tank tests are also illustrated. The establishment of underwater dynamics model provides the operating basis for an accurate control in the target location experiments.Secondly, the underwater proximate sensor for multi-targets positioning system is studied. Different from the geometric inter-potation method proposed before, a novel multi-targets positioning method based neural network is illustrated. This method has an advantage of no limitation on target number. Two different simulation experiments are described. The simulation results of regional positioning and accurate angle positioning show the efficiency and accuracy of the method.And then, underwater visual targeting system is described. The detail principle and composition of underwater vision systems are presented. The paper also presents the integration solutions and improves the generalization ability method. According to different shapes based on the goals or objectives in the image location area; simplify the structure of each sub-network and improve the generalization ability of the system. The experiments result show the sufficient accuracy output can be also gained for those objects which are not in the training sample. Meanwhile, keep adding the sub-neural network and perfect the system gradually, the targeting ability of the system is improvedFinally, an integrated control system for underwater multi-targets is established. Integrate the proximate sensor and visual camera information and complete the output fusion based the LVQ neural network. For proximate sensor, the positioning methods based on neural network can offer an accurate angle information while it can be better adapt to the underwater strongly nonlinear, strong noise environment in terms of the visual camera based on neural network-based positioning methods. With the nerve network integration technology, each single neural network can be divided into multiple subnets. Thereby, the number of training samples can be reduced for each sub-network and the generalization ability of neural networks will be boosted. And the high algorithm complexity problem is also solved.In this paper, a series of studies and experiments on multi-targets location based on multi-sensors are carried out combined the relative requirement of the National Research Project. The results provide a theoretical basis and technical means for underwater robots self-operating positioning control system.Meanwhile, a lot of work needs to be further researched.