Crowd Size Estimation using Image Processing

Abstract

Crowd analysis studies the behavior, movement, and patterns within a group of people newlineusing computer vision and machine learning techniques for crowd security. In recent newlineyears need for crowd analysis that can estimate crowd size, has arisen to manage the newlinecrowd well. Crowd analysis can help in crowd management which is very important in newlinevarious situation such as crowd at big events, traffic management, providing security to newlinecrowd. Crowd analysis involves two kinds of analysis, (1) Crowd density estimation, newlinein which we estimate the number of people in a given area or crowd, and (2) Crowd newlinecounting methods, in which we detect and count individuals within frames or images, newlinewhich can be beneficial for crowd management and safety. Thesis focuses on a convolutional newlineneural network based approach for crowd estimation which incorporates the newlinephysical distance information that can help in deciding if the crowd density at a particular newlineplace is safe or unsafe. Also mask wearing status helps us where specific instructions newlineare needed with top priority, in case of any symptoms of infectious disease due to less newlineproximities are found. Crowd estimation is an important and challenging problem because newlineof its importance in several security related applications. The proposed model is newlineevaluated on the standard dataset. The experimental results show the effectiveness of the newlineproposed model. Thesis also focuses on facial emotional recognition for a crowded image. newlineThe success of this method can be attributed to the integration of proposed method newlinewith crowd density estimation methods to get the accurate emotion of the people in a newlinecrowded area. It is useful, whenever a crowd gather at a particular place and there is a newlinexvii newlineneed to get the people emotion, to avoid any line of tension/crying/shouting. It can take newlinecognizance for overcoming possible mishappening. The outcome of this research not newlineonly enhances crowd analysis techniques but establishes a basis for developing more newlineresilient crowd analysis systems with broader applicability in domains