Crowd Detection in Still Images Using Combined HOG and SIFT Feature

Abstract

Person detection and tracking in crowd is a challenging task. We detect the head region and based on this head region we can detect people from crowd. Individual object detection has been improved significantly in recent times but the crowd detection and tracking contains some challenges. Crowd analysis is a highly focused area for law enforcement, urban engineering and traffic management.  There are a lot of incident occurred in crowd area during some fabulous event. In this research low resolution and verities of image orientation is a key factor as well as overlapping person images in crowd misguided the system. An enhanced system of interest point detection based on gradient orientation information as well as improved feature extraction HOG is used for identifying the human head or face from crowd. We have analyzed different types of images in different varieties and found accuracy 88-90%. In a number of applications, such as document analysis and some industrial machine vision tasks, binary images can be used as the input to algorithms that perform useful tasks. These algorithms can handle tasks ranging from very simple counting tasks to much more complex recognition, localization, and inspection tasks. Thus by studying binary image analysis before going on to gray-tone and color images, one can gain insight into the entire image analysis process.