Research On Surveillance Target Classification For Video Surveillance System

The construction of public security video surveillance system have strict requirements and detailed considerations.As a technical facility for social security prevention,the front-end point of video surveillance system is always built around agency units,commercial facilities and densely populated public areas.The location of these surveillance cameras can be called static control targets for video surveillance systems.POI is the map of all kinds of enterprises and institutions in the urban geographic space,which has the characteristics of accurate information and good real-time performance,and the name of which can largely reflect the basic properties of POI functions and types.From the POI information,it is an important and feasible method to preferentially consider the key surveillance targets of video surveillance front-end point.From the perspective of text length,the POI names are short and the feature words are sparse,so it is difficult to classify them through traditional methods.From the perspective of classification,the more categories of targets,the worse of the classification results.The purpose of this paper is selecting the key surveillance targets of public security video surveillance system through the text classification of POI names,including the target categories classification,the text representation model,text classification model and so on.The specific work is as follows:In terms of the classification criteria of the surveillance targets,according to the public security video surveillance system standards of the front-end point position and the the regulations of the internal security work in the enterprises and institutions,this article has reclassified the surveillance targets categories into three major categories and twenty secondary categories.The first major category is the surveillance targets in important areas of the public security’s video surveillance system,the second one is the targets which should be set up acquisition points by public security agencies,and the third one is the targets which can be set up acquisition points by social organizations.In terms of the classification algorithm implementation,according to the text features and classification requirements of the surveillance targets of video surveillance system,combining the weight calculation method of TF-IDF and CHI,the text representation methods based on Word2 vec model are improved,which are W2V-TFIDF and CW2 V.Besides,combining with the Support Vector Machine,the classification models for the surveillance targets of video surveillance system are built.The simulation experimental results show that,compared with traditional text classification models,the accuracy of the classification model proposedhas been significantly improved and the accuracy of CW2V-SVM model is the best,of which macro average F1 value is 0.9383 when the dimension of word vector is 250.In addition,the visual interface of the surveillance targets classification is designed and realized,and the classification results are displayed intuitively which have certain reference significance for the selection of the front-end point of the video surveillance system.