Cross camera people counting with perspective estimation and occlusion handling

We introduce a novel approach to cross camera people counting that can adapt itself to a new environment without the need of manual inspection. The proposed counting model is composed of a pair of collaborative Gaussian processes (GP), which are respectively designed to count people by taking the visible and occluded parts into account. While the first GP exploits multiple visual features to result in better accuracy, the second GP instead investigates the conflicts among these features to recover the underestimate caused by occlusions. Our contributions are threefold. First, we establish a cross camera people counting system that can facilitate forensics investigation and security preservation. Second, a principled way is proposed to estimate the degree of occlusions. Third, our system is comprehensively evaluated on two benchmark datasets. The promising performance demonstrates the effectiveness of our system.