Abstract
The crowd estimation presents the main problem of many applications as a pedestrian crowd, traffic flow. The crowd estimation based on computer vision is faced to scale variation, occlusions, and non-uniform distribution challenges. This defy could be seen as a physical problem by considering it as a particle dynamics-model or as a computer-vision model. Seen that importance an accurate estimation provides in-depth the behavior of the crowd. We propose a smart crowd estimation system based on video processing. The proposed uses the k-means clustering and the Balanced Communication-Avoiding Support Vector Machine classifier to estimate the number of pedestrians in crowds. The achieved results prove the accuracy of the system in comparison with previous works. The error is less than one for a low-dense crowd, around two eighty-eight of persons for a medium-dense crowd, and around eighty-nine persons for a highdense crowd. Findings not only show a low error of the density but also a minimum time of execution (13 ms).