IntenCT: Efficient multi-target counting and tracking by binary proximity sensors

Binary proximity sensors (BPS) is a generic model for many non- collaborative, presence detecting sensor. It outputs ''1'' when one or more targets are presenting in its sensing range and ''0" otherwise. It cannot tell the number of targets nor the targets' identities in its sensing range. But for its privacy protection and device-free properties, BPS-based tracking has attracted great attentions. However, multiple target counting and tracking (MTCT) by BPS network remains very challenging. Existing approaches generally rely on trajectory decomposition, which suffer association complexity issue and can hardly provide accurate results. To address these challenges, this paper presents an novel intensity-based counting and tracking approach, called IntenCT, which tracks the evolvement of the multi-targets' probabilistic density distribution overtime, without the complexity of enumerating the multiple targets' trajectories. Then, clustering algorithms on the density distribution are proposed to find the target groups, and count the targets in each group by calculating the integral of the density distribution in the group region. At last, the trajectories of the separable targets in each group are estimated using K-means and a motion consistency model. Extensive analysis and simulations show that IntenCT has quadratic complexity which is very efficient; provides the current best known multi-target counting lower bound; and tracks the multi-targets more accurately than the existing approaches.