Abstract

In bad weather, the visibility at sea is low, and it is difficult to rescue people and ships in distress. Existing sensors such as visible light cameras and LIDAR will fail in bad weather, so this paper proposes a design scheme for maritime emergency rescue based on thermal infrared cameras. Infrared thermal imaging is a technology that converts infrared thermal radiation into electrical signals and displays them through the screen, which is unaffected in low visibility environments such as bad weather and night and still works normally, and is a great aid to maritime emergency rescue. The accurate identification of maritime vessels is achieved by selecting the YOLOV5 target detection algorithm with high accuracy, the accurate ranging of maritime vessels by monocular ranging algorithm, and finally the whole project is deployed to NVIDIA Jetson AGX Xavier high-performance edge computing device for inference acceleration using TensorRT framework. To verify the effect of different sizes of network structures on the training of the dataset, experiments were conducted on YOLOV5’s s-model, m-model, l-model and x-model, respectively, and the results showed that the detection results of these four models were comparable, however, the inference frame rates did vary greatly. In windows the inference frame rate is 19fps for the s-model, 15fps for the m-model, 9fps for the l-model and 4fps for the x-model. The s-model was chosen to be used in the subsequent study. The s-model was accelerated by TensorRT to increase the frame rate from 19fps to 30fps, enabling real-time detection and ranging functions. The experimental analysis shows that the scheme of this paper is feasible and reliable.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call