Research On Human Detection Technology In Fishing Boat Cockpit Video Based On Embedded GPU

The fishing boat cockpit is a core part of the safe navigation of fishing boats.The safety accidents of many fishing boats are caused by illegal operation of the cockpit.Therefore,the use of video surveillance technology to conduct human detection and population counting of fishing boat cockpit operators is very important to the safety of the entire fishing boat.The embedded platform has the characteristics of small size,low power consumption and easy deployment.It is suitable for the problem that the space inside the fishing boat cockpit is small and the shipboard electronic equipment needs to meet the requirements of waterproof,anti-corrosion and anti-salt fog.This paper chooses embedded platform as the hardware platform for fishing boat cockpit video detection.However,since video processing requires a large number of operations,the performance requirements of the hardware processor is also high.The computing performance of the ordinary embedded platform cannot meet the requirements,and the embedded GPU with low power consumption and high performance becomes an effective solution to this problem.The main work done in this paper is as follows:(1)Analyze the problems existing in the human detection module of the existing fishing boat safety monitoring system.For the problem that the existing fishing boat cockpit detection system based on OpenCV and HOG features is not accurate and the detection speed is slow,this paper proposes idea of video detection for fishing boat cockpit human based on embedded GPU.(2)An embedded fishing boat human video detection system based on YOLO object detection framework was built.In order to improve the accuracy of system detection,the special scene of fishing boat cockpit was modeled and the special data set of fishing boat cockpit was made.The system is implemented on the embedded processing platform NVIDIA Jetson TX1.The test results show that the system improves the detection accuracy by 10% and the detection speed by nearly 30 times compared with the existing fishing boat cockpit human detection system.However,the power consumption of the system is also greatly increased,and the long-time operation requires the support of cooling fan,which Unable to fully adapt to the requirements of a no fan system in a fishing boat cockpit environment.In addition,the hardware cost of the system is also greatly increased.Considering that the deep learning-based YOLO object detection framework implementation model is more complex and limited by the hardware platform specified by CUDA.Select the DPM algorithm with the same computational core as the convolutional neural network representation.(3)Building a low-power fishing boat human video detection system based on embedded GPU parallelized DPM human detection algorithm.The DPM algorithm is fully theoretically researched,and the algorithm is transplanted in embedded Linux.The time-consuming analysis and parallel analysis of the algorithm are performed to determine the bottleneck module for limiting the detection speed of the algorithm.The cross-platform OpenCL isomerism is utilized.The programming model optimizes the parallelism of the algorithm and improves the detection efficiency of the algorithm.On the basis of the fishing boat cockpit special data set,a multi-component model dedicated to the detection of the cockpit human of the fishing boat was trained to ensure the detection accuracy of the algorithm.The implementation of the system was completed on the embedded platform Firefly RK3288.The experiment results show that,through performance testing and comparative analysis,compared with the existing fishing boat cockpit human video detection system based on OpenCV and HOG feature,the detection accuracy of the optimized DPM algorithm can be improved by more than 10%,and the detection speed can be increased by 7 times.Compared with the detection system based on the original DPM algorithm,the detection speed is increased by 4 times while the detection effect is guaranteed.Compared with the power consumption of the embedded fishing boat cockpit human video detection system based on the YOLO object detection framework,the detection speed is slightly less,but the power consumption is reduced to less than 7W,which can realize the continuous operation in the fishing boat cockpit without fans,and the hardware cost is also greatly reduced.Comprehensive final inspection accuracy,speed,power consumption and cost considerations,the low-power fishing boat cockpit human video detection system based on embedded GPU and parallelized DPM algorithm proposed in this paper is a personnel detection system that can balance low cost,low power consumption and high performance.It has certain reference significance for the application of personnel detection technology in special application scenarios.