Research On Detection Technique Of Unusual Behavior Of Workshop Workers Based On FPGA

With the development of neural network learning algorithm and computer hardware,the advantages in large-scale image processing are becoming more and more obvious.However,due to the huge amount of computation and storage requirements,but also brings great difficulties for the implementation of neural network hardware.Because FPGA has the advantages of low power consumption and high parallelism,it is suitable for the implementation of neural network acceleration.In order to meet the requirement of high speed and low power consumption of workshop application,a high precision,high speed and low power consumption detection system for workshop workers is developed based on the platform of ZYNQ-7020 series,which combining the advantages of FPGA and ARM.In this paper,a hardware acceleration scheme based on FPGA is proposed,which combines ARM + FPGA to realize the design of hardware and software collaboration,so as to detect and deal with workers’ abnormal behavior effectively.The main work of this paper is as follows:1)A model for detecting abnormal behavior of shop workers based on improved TSN is designed to solve the problem that traditional CNN cannot extract characteristic information from time series and that traditional denser optical flow method has some errors in distinguishing similar actions.The model not only improves the feature information extraction in the time series,eliminates the influence of complex scenes on the recognition of workshop workers’ abnormal behavior,obtains more precise contours of moving targets,and improves the accuracy of workshop workers’ abnormal behavior detection.2)How to make the neural network algorithm and hardware equipment cooperate.In order to better meet the needs of daily life and improve the efficiency and real-time of deep learning algorithm,neural network algorithm works with hardware devices.At first,the ZYNQ-7020 platform is designed as the core module of the hardware of the worker abnormal behavior detection system,and then other peripheral circuit modules are added according to the function of the system.By accelerating and optimizing the design of TSN network at all levels,the efficiency of operation can be effectively improved,and the network model can be combined with hardware,which can effectively reduce the operation cost of the system.3)The advantages of the system are proved by the PYNQ framework in terms of function,stability and real-time.The results show that the FPGA + ARM isomeric system designed in this paper can detect abnormal behavior of shop workers and achieve an acceleration ratio of39.58 times at a working frequency of 150 MHz of up to 30 frames/s compared to ARM implementations in pure software.In addition,the average power consumption of the system is only 2.898 W,which satisfies the design requirements of real-time,accurate and high performance video surveillance system.