Counting People Based on Linear, Weighted, and Local Random Forests

Recently, many works have been published for counting people. However, when being applied to real-world train station videos, they have exposed many limitations due to problems such as low resolution, heavy occlusion, various density levels and perspective distortions. In this paper, following the recent trend of regression-based density estimation, we present a linear regression approach based on local Random Forests for counting either standing or moving people on station platforms. By dividing each frame into sub-windows and extracting features with ground truth densities as well as learned weights, we perform a linear transformation for counting people to overcome the perspective problems of the existing patch-based approaches. We present improvements against several recent baselines on the UCSD dataset and a dataset of CCTV videos taken from a train station. We also show improvements in speed compared with the state-of-the-art models based on detection and Deep Learning.