A PTZ camera based people-occupancy estimation system (PCBPOES)

摘要

A single PTZ Camera Based People-Occupancy Estimation System (PCBPOES) is proposed to estimate the number of people occupying a region of interest with acceptable accuracy. The PTZ camera aids this objective by efficiently monitoring a wide area by dividing it into zones, capturing high resolution zone images aided by the optical zoom for detecting human head patterns. A seminar room is used as a test location for evaluating this system. Cascade trained support vector machine models are used to detect head patterns in the images. Features such as HAAR, HOG were explored for modeling the head patterns. We adopt a zone-based detection and ensemble approach to combine the zone detection results. A temporal probabilistic selection approach improves the head detection accuracy. Experiments prove that the system has high detection accuracy achieving average precision scores of 0.92 and recall scores of 0.82. A bio-inspired retina model is adopted to detect heads in low illumination. The system is currently used to estimate people-occupancy in the test location.