Detection-based tracking for crowded targets in distributed visual sensor networks

Visual sensor networks (VSNs) have generated a new emerging interdisciplinary research field and got attentions from many diverse research disciplines due to their potentials to solve problems in multi-camera applications. Since each camera node has limited processing, sensing, energy, and bandwidth capabilities, collaboration in sensor networks is required not only to compensate the limitations of nodes but also to improve the accuracy and robustness of the network. In this paper, we present collaborative target detection and tracking algorithm for crowded targets in VSNs, a challenging problem because of the extremely higher data rate and the directional sensing characteristics of cameras. In traditional detection-based tracking algorithms, targets are detected at the intersections of the back-projected 2D cones of each target generated at different sensor nodes. However, the existence of visual occlusions would generate many false alarms. In our approach, instead of resolving the uncertainty about target existence at the intersections, we identify and study the non-occupied areas in 2D cones and generate the so-called certainty map of non-existence of targets. Additionally, we propose a dynamic itinerary for progressive certainty map integration where a limited amount of data is transmitted among only a limited number of nodes. When the confidence of the certainty map is satisfied, targets are detected at the unresolved regions on the certainty map and a Gaussian-based motion model is applied to track targets. Based on the results from real experiments, the proposed distributed method shows effectiveness in tracking accuracy.