Research On Intelligent Video Surveillance Technology Based On PYNQ

As the demand for fire in daily life increases,urban density increases,and people’s awareness of fire prevention is inadequate,the drawbacks of traditional fire detection sensors are becoming more and more significant,and people are beginning to think about and seek more efficient,stable,and applicable fire warning systems.The rapid development of deep learning,high-definition video surveillance,computer vision technology and the improvement of computer computing power have provided the technical basis for new fire prevention and control systems,and people have started to try to apply these emerging technologies to fire prevention and control by processing and analyzing the information of various features in the surveillance images to achieve the recognition and early warning of fire.In recent years,deep learning-based target detection algorithms have emerged,and the efficiency of accurate target recognition and detection has become increasingly high.However,deep learning also has the disadvantage that the required computer computing power resources are large.With more and more intelligent devices connected to the Internet,the current computing power of existing computer equipment is difficult to support the computational needs of the massive amount of data generated every day,and the real-time data transmission of various systems is not guaranteed due to arithmetic overload,resulting in a higher rate of missed detection and false detection of target detection.At the same time,some systems rely too much on expensive GPU hardware platforms and devices,and the excessive cost also constrains the growing intelligent needs of society.As a new type of computing based on the idea of distributed architecture,edge computing can effectively relieve the pressure and challenges faced by cloud servers in handling massive data.In this paper,we conduct an in-depth study on the technology of intelligent video surveillance system for fire scenario,and combine the target detection algorithm based on deep learning framework with edge computing technology to implement an intelligent video surveillance system based on edge computing,and apply it to industrial field test.The main work of this research is as follows.(1)Proposed improved differential algorithm to extract motion targets.Combining the inter-frame differencing method with the background subtraction method to achieve complementary advantages,and then combining the sliding average filtering algorithm to achieve adaptive updating of the background,the improved differencing algorithm is applied to image pre-processing of the surveillance screen to extract the areas suspected of flame and smoke motion in the screen,and improve the accuracy of the subsequent system for flame and smoke target detection.(2)Flame and smoke target detection based on deep learning.By collecting pictures and video data of various fire scenes in practice,a custom fire dataset is made and the dataset is expanded by using data augmentation.At the same time,the anchor boxes are re-clustered according to the custom dataset to improve the detection effect of the model on flame and smoke targets of different sizes.(3)Build a PYNQ-Z2-based video surveillance system.The improved differential detection algorithm and deep learning target detection model are deployed to the PYNQ-Z2 development board,and the PYNQ-Z2-based video surveillance system is used to collect and process the images monitored by the network cameras in real time and realize the intermodulation operation of edge computing and server side.The test results show that the system can achieve the goal of real-time warning of fire at industrial sites.At the same time,in order to improve the portability of the system designed in this project,different edge computing development boards are used for comparison experiments.This research combines edge computing,computer vision technology and deep learning target detection technology to implement an intelligent video surveillance system that can be applied to various security scenarios to achieve hazard warning.