Pedestrian crowd level estimation by Head detection using bio-inspired retina model

A system to estimate pedestrian crowd levels is proposed. It uses the Parvo channel output of the bio-inspired retina model for improved sensitivity to head patterns in low illumination. Head features are learned from the parvo output. Several features are explored, namely, Aggregate Channel Features (ACF), Integral Channel Features (ICF) and the Histogram of Oriented Gradients (HOG). ACF was selected based on the experimental results. The system is capable of detecting heads in most of the possible head poses, in low illumination situations also. A probabilistic temporal selection approach is proposed to improve the accuracy of the head detection and thereby improving the crowd level estimation accuracy. The system is tested with several standard research data sets such as the IIT Head pose and IHDP Head pose data sets. It exhibited superior performance with average precision and recall scores of 0.99 and 0.97 respectively. The system is currently used to estimate crowd levels in public locations like canteen seating areas.