An Efficient Feature-based Method for People Counting
Daniel Helmer, Heiko Hinkelmann, Thomas Hollstein · 2023
This paper presents a novel feature-based method for detecting and counting people in smart building applications using Time-Of-Flight (TOF) cameras. A TOF camera images a scene three-dimensionally and can detect people passing through below the camera in a privacy-compliant manner. The paper shows the researched image processing chain for people counting, which uses feature extraction for the robust classification of persons and non-person objects. 34 feature candidates have been developed, tested, and statistically evaluated regarding their suitability for classification. Of these 34 candidates, the three best-suited features were selected for the final solution. Experimental results of different use cases show that 99% of 898 classifications were performed correctly. Among 600 counting events, 99% of the use cases with one or two persons walking separately were counted correctly, while less accurate counting was observed when two people were closely walking side by side. The low computational effort of this method allows efficient embedded system implementations.