Development of an Improved Technique for Analysis of Crowd Behavior

Abstract

Behavior analysis of crowd is a thought-provoking job because of the variation of density and scene inconstancy. This research plans to address various difficulties in picture as well as recordings of various crowd systems. The principle objective is to assist in quantizing the group conduct. This research focuses on developing better techniques for clear understanding of crowd behavior in various phases. An improved technique is developed using a new sky detector to detect and extract the sky region from images to avoid inaccuracy and improve the descriptor results to analyze the behavior of crowd. An image is divided into two parts i-e sky region and ground region using the sky detector followed by the extraction of the sky area from the data. A KLT tracker is used to track the crowd. Coherent Detection is performed for clustering. Crowd motion is also analyzed with novel descriptors i-e speed, direction, shapes and merging probabilities to get satisfying results. Different scenes fromdifferent places are taken for the experiments to perform descriptors as shape, speed, direction and merging probabilities. The whole methodology is capable of analyzing the crowd behavior effectively without involving the sky region and provides a befitting accuracy in results. It generates satisfied results according to the different scenes. This research work concluded that by extracting the sky region using sky detection technique is more impactful rather than considering the clusters on ground and in sky too. The proposed method and results show a befitting accuracy as compared to the previous research.