Research And Application Of Surveillance Video Target Detection And Tracking Based On OpenCV

With the rapid development of information technology,people have higher and higher requirements for security.In such an environment,a monitoring system comes into being.The popularity of cameras has prompted the continuous development of technologies in the field of surveillance,and intelligent surveillance systems have become possible and widely used.Video target detection and tracking technology is the core of the intelligent monitoring system,and also a key technology in the fields of computer vision,image processing,artificial intelligence and so on.At present,the intelligent monitoring system is mainly devoted to achieving autonomous detection and tracking of targets.Therefore,this paper studies and implements the target detection technology and tracking technology based on deep learning.This article is based on the Open CV computer vision library and the programming language is Python.In the field of target detection,the SSD algorithm is selected as the research object.The SSD algorithm draws on the anchor mechanism of Faster RCNN and the regression strategy of the YOLO algorithm,and has good detection accuracy and real-time performance.However,in order to pursue better real-time performance,we use the lightweight network Mobile Net to replace the traditional SSD basic network VGG16,and compare the performance of the Mobile Net V1?SSD algorithm,Mobile Net V2?SSD algorithm and VGG16?SSD.Experiments show that the real-time detection of Mobile Net V2?SSD is much higher than the traditional SSD algorithm.Aiming at the detection of pedestrians in surveillance videos,a pedestrian data set is established for migration training to obtain a pedestrian detection model and test experiments.Aiming at pedestrian tracking in surveillance video,this paper compares the classical kernel correlation filtering algorithm on otb2013 data set,selects some data sets with occlusion attribute to show the accuracy.From the experimental results,KCF algorithm has high real-time performance and accuracy,so kernel correlation filtering algorithm is used for tracking.However,in view of the shortcomings of the kernel correlation filter algorithm's poor robustness in more complex situations such as occlusion deformation,this paper uses a convolutional neural network to extract features from the image,integrates the extracted features into the tracking algorithm,and monitors the sequence in the surveillance video sequence.The method was verified by experiment.Experiments prove that the KCF algorithm using deep features is more effective in pedestrian video sequences than the KCF algorithm using HOG features.