Two Step Framework on Indoor Occupant Counting

Knowing the number of occupants of different zones in buildings, is crucial for intelligent buildings and building energy managing system. In this paper, a two-step framework on indoor occupant counting is proposed, where the region of interest (ROI) of the target image is obtained based on a cascade classifier in the first step, and the number of indoor occupant is estimated through classifying and recognizing the ROI area based on deep learning technologies in the second step. Finally, this system framework have been implemented and real experiments are conducted to verify its effectiveness. Results show that the average of more than 91% accuracy rate on the occupant number detection can be achieved by the partly background elimination, and the binary absence or presence prediction accuracy can reach more than 93%.