Pedestrian density estimation system using Time-Spatial Image (TSI) processing and short-term motion vector

Abstract

Estimation of pedestrian density and/or area occupation on a real-time basis is quite challenging to be implemented automatically, economically and accurately. The traditional data collection approach is laborintensive and time-consuming. Alternative image processing approaches required a high computational resource to extract and track pedestrians accurately. This paper introduces a real-time area occupation system using Time-Spatial Image (TSI) processing and short-term motion vector which can be performed on the realtime basis without using large amounts of processing resources from pedestrian extraction and tracking. TSI is an image of numerous lines against time. The proposed system will estimate TSI from a virtual detection line. After a short time period, the detection lines can be constructed as TSI. In this research, the camera are installed in the observation area on a gantry with a top-down view, 90 degrees to the horizontal axis, to avoid the pedestrian privacy issue. The proposed system will estimate multiple TSIs at the same period from different virtual detection lines location within the observed location. The study exploit the attributed of the direction of pedestrian height within the TSI to estimate the short-term individual pedestrian direction so called short-term motion vector. The proposed system pedestrian density result will be validated with the density estimated from perspective transformation. The results are shown as pedestrian density in term of area occupation based on the inflow and outflow at the observed location.