Audio-Video based People Counting and Security Framework for Traffic Crossings

摘要

In this paper, we propose a traffic management system which uses both audio and video data at intersections to prevent traffic congestion. A novel scheme of counting and tracking crowd with a single overhead camera is proposed for the purpose of real-time traffic control at crossings as well as for adjusting the carrier frequency using the video data. The system recognizes people movement along various directions estimating the possibility of traffic congestion. To carry out the tracking procedure, various temporal and spatial features of the images are used to identify the people in the crowd in order to predict the position of the objects in the current frame. Several issues such as emergence of people, departure of people, occlusions, and de-occlusions are resolved through interactions between regions and objects. With the help of audio data at counting region, we identify children by their voice using pitch analysis. The database constitutes of voices of speakers of all ages and of both genders. The paper also includes automated accident detection at the traffic intersections through use of audio data at intersections. It can classify the sounds into "crash", "power brake" and "normal traffic" sounds. The experimental results show the effectiveness of our framework.