Crowd Counting Using Group Tracking and Local Features

Abstract

In public venues, crowd size is a key indicator of crowdsafety and stability. In this paper we propose a crowd count-ing algorithm that uses tracking and local features to countthe number of people in each group as represented by a fore-ground blob segment, so that the total crowd estimate is thesum of the group sizes. Tracking is employed to improve therobustness of the estimate, by analysing the history of eachgroup, including splitting and merging events. A simpli-fied ground truth annotation strategy results in an approachwith minimal setup requirements that is highly accurate.