Particles cross-influence for entity grouping

In this paper we propose a novel approach to detect and track moving entities in wide surveillance video. Considering the wide area covered by the camera, which makes the detection and tracking of humans, as well as the classification of their motion a complex task and resource consuming, we adopt a particle-based approach to highlight particles of interest and group them based on their motion properties. A cross-influence matrix is computed at the particle level identifying the relevant areas of the video, and pruning static particles and outliers. Based on the motion features of the particles marked as interacting with their neighbors, a learning procedure based on an MLP neural network is implemented, in order to create consistent groups, representing the moving entities to be tracked over time. The method has been tested on two publicly available datasets with different resolutions and motion characteristics.