Multi-Attribute NMS: An Enhanced Non-Maximum Suppression Algorithm for Pedestrian Detection in Crowded Scenes

Removing duplicate proposals is a critical process in pedestrian detection, and usually performed via Non-Maximum Suppression (NMS); however, crowded scenes, the detection of occluded pedestrians are hard to distinguish from proposals, making results inaccurate. In order address above-mentioned problem, authors this paper propose Multi-Attribute NMS (MA-NMS) algorithm, which combines density count attributes adaptively adjust suppression, effectively preserving while removing proposals. obtain attributes, an attribute branch (ATTB), uses context extraction module (CEM) extract pedestrians, then, concatenates with features predict both simultaneously, also proposed. With proposed ATTB, detector, based on MA-NMS, constructed for scenes. Extensive experiments conducted using CrowdHuman CityPersons datasets, show that method outperforms mainstream methods AP (average precision), Recall, MR−2 (log-average miss rate), sufficiently validating effectiveness MA-NMS algorithm.