Safety for pedestrian recognition in sensor networks based on visual compressive sensing and adaptive prediction clustering

Abstract Aiming at the imbalance between energy use and tracking accuracy in multi-sensor target recognition, a pedestrian target recognition method based on visual compressed sensing and adaptive predictive clustering is proposed to track multiple pedestrians simultaneously. After acquiring the pedestrian target image, the scale invariant features of the pedestrian face in the image are extracted firstly, and the target is sparsely represented by the feature dictionary. Then adaptive prediction clustering is used to capture the change of pedestrian behavior attributes. Then, the sensor is selected by Region method, and the sensor contributing to the pedestrian area is activated to realize the pedestrian tracking. In the simulation scenario, 500 sensors are randomly deployed in a given square area. Because of fewer sensors and shorter computation time, the network lifetime has been significantly improved.