Research And Design Of Virtual Electronic Fence System Based On Video Surveillance

In daily production and life,accidents of falling objects often happen.In some construction sites or other areas where objects may fall,fences are often set to prevent people or pets from entering the marked dangerous areas.However,the traditional fence has some problems,such as less interception power,not timely response to intrusion alarm and so on.Because of its rich content,wide coverage and strong real-time,video monitoring has become an important tool to protect the safety of people's lives and property and maintain social security.The main work of this paper is as follows:(1)The functions of video image acquisition,jitter removal and ROI selection are realized.Based on the pre-processing of image grayscale and binarization,the de jitter processing calculates the projection of the selected reference frame,and calculates the correlation coefficient of the video reference frame and each subsequent frame,so as to achieve the purpose of removing jitter and maximizing the selected area of the video image.The function of ROI selection can be realized by drawing any polygon in the monitoring image with the mouse,taking the selected ROI boundary as the virtual fence boundary,and only detecting and recognizing the image entering the ROI.(2)The target detection method based on yolov3 algorithm is studied.The algorithm principle and training process of yolov3 are studied in depth.According to the research background,the image types of dataset are set,which are divided into people and pets.According to the actual situation,the image types of forbidden risk areas are divided into seven categories: people,bicycles,cars,motorcycles,trucks,cats and dogs.Yolov3 model is used for training,target detection and detection results are recorded Fruit.Finally,the detection results of 150 target images of seven categories are analyzed by calculating the evaluation parameters.The results show that the recall rate and accuracy rate of all kinds of images are higher than 90%.(3)The target detection method based on retinaet is studied.In this paper,the algorithm principle of retinanet model is studied deeply,the network structure,loss function and anchor mechanism of the model are studied and explored,according to the training steps of the model,the same image type test model as yolov3 model is used,the test results are recorded and the evaluation parameters are calculated,compared with the test results of yolov3,and the detection accuracy and anchor mechanism of the two target detection models are compared Applicable image types.(4)Based on the research of target detection algorithm,the virtual electronic fence system based on video monitoring is designed and implemented,and the software and hardware platform of the system is built.The test results of the system show that the detection accuracy of seven categories of images set for ROI area detection and recognition is high,and the virtual electronic fence function based on video monitoring researched in this paper is realized.In a word,this paper studies and designs a set of virtual electronic fence system based on video monitoring.Compared with the traditional fence,this virtual electronic fence makes up for the defect that the traditional fence only has a single fence function,and increases the function of real-time online monitoring and alarm deterrence.At the same time,it improves the problems of too many mainstream electronic fence hardware devices and complex laying process,and simplifies the laying process The installation and laying process reduces the cost and operation difficulty.The result of the test shows that the pedestrians,cars,motorcycles and bicycles in the video can be correctly detected,the accuracy and recall rate are higher than 90%,and there is no missing or wrong inspection.