Counting people using gradient boosted trees
Bingyin Zhou, Ming Lu, Yonggang Wang · 2016 IEEE Information Technology, Networking, Electronic and Automation Control Conference · 2016
This paper proposes a real-time approach to count the people in crowded scenes using holistic properties of the video, without using individual detection or tracking. A group of efficient holistic features are firstly extracted from the crowd segmentations. Furthermore, a nonlinear function mapping the feature vector into the number of people is learned with gradient boosted trees. Experiments demonstrate that our method can perform more accurate people counting.