Video-based crowd counting with information entropy

As a key indicator of safety, the number of persons in pubic venues is quite important. However, most algorithms require a burdensome training, which is far away from practical application. In this work, we introduce a counting approach with information entropy (IE). Without extracting features or tracking objects, this algorithm greatly simplifies the process of counting. Firstly, the moving objects are segmented by background subtraction. And then interested targets are normalized to avoid perspective effect. Finally, we compute the IE of the normalized images. In theory, the IE is proved to be approximately linear with the number of persons. However, considering the deviation from occlusion, perspective distortion, difference between pedestrians etc., we also make quadratic fitting for higher accuracy. The experimental results show that the accuracy of pedestrian number obtained by IE algorithm is higher than that of the previous research. So, the usage of IE in this field is efficient and practical.