Research And Application Of Intelligent Video Surveillance Technology Based On Machine Learning

In recent years,the rapid development of 5G,computer hardware,artificial intelligence and other technologies has brought more possibilities for video surveillance technology.Traditional video monitoring system relying on artificial is changing to intelligent video monitoring system.The intelligent transformation of the surveillance system is leading a new round of development in related industries,and more relevant universities and enterprises are racing to invest in the development of intelligent video surveillance systems.The core of intelligent monitoring system lies in the analysis and understanding of image and video content.Benefiting from the strong potential of deep learning,related technologies are experiencing rapid progress.This article takes the general road environment as an application scenario,explores and develops a reliable intelligent video surveillance system that meets the needs of practical applications.The system realizes the dynamic tracking and alarming of the target in the video sequence.Firstly,the pedestrian detection technology based on traditional machine learning is explored,and genetic algorithm is introduced to optimize the core parameters of the classifier,so as to solve the problem that the classifier is greatly affected by the parameters;at the same time,the adaptive enhancement algorithm is used to significantly improve the classification ability of the classifier.Experimental testing and analysis show that the improved model has better detection effect and can be applied to solve the problem of pedestrian detection in scenarios with low real-time requirements.Then,an improved convolutional neural network model based on yolov3 is proposed for the application scenarios with more target categories and more complex environments.On the basis of the former,combined with the characteristics of the application scenario,the model is improved by the adaptive adjustment of anchor frame mechanism,multi-scale detection improvement,embedded senet block and loss function optimization,which improves the detection ability of the model.Then using the improved detection algorithm,a multi-target tracking method based on detection is proposed.This paper solves the problem of target association between observation target and tracking track,and gives the solution to the problem of target disappearance,new target appearance and error matching.Finally,an intelligent video monitoring system is designed based on the multi-target tracking method,which can automatically detect,locate and track multi-category targets in general security scenarios,and complete the detection and alarm task for specific category targets.