Research And Application Of High-Density Crowd Counting Method

The social significance and market value of high-density crowd counting are huge. With the information about the number of people in a ROI area, manager can achieve the optimal schedule of manpower and resources. For some public place like square and passage, the estimation of the number of people present in those places can be extremely useful to early warning. Thence, several works in the fields of video analysis and intelligent video surveillance have focused on this issue.In the traditional approach, people in the scene are first individually detected, using some form of segmentation and object detection, tracked and then counted. While traditional approach has a high requirement on video resolution and is just suite for low-density crowd, would be incapable for high-density crowd and open environment. By Comparing research trends and methods of this issue in domestic and foreign, we presents a SURF-based method for high-density crowd counting, focusing on overlaying the low counting accuracy in a high-density crowd or open environment. The main content and innovation of the proposed method is included in the following categories:(1) Detecting the interest points associated to people by using a block-matching technique. On other hand, we propose a new clustering algorithm MST-DBSCAN, which is derived from the traditional density-based clustering algorithm (DBSCAN) by adopting minimum spanning tree (MST), making its minimal search domain adaptive to the distribution of clustering data.(2) Making the number of interest points to be the main character of crowd eigenvectors. Meanwhile, in order to adapt to high-density crowd and open environment, we take the effect of perspective distortion on the number of detected interest points into consideration. (3) With support vector regression, we make the training of crowd eigenvectors and get the prediction model. Finally, we do estimation the number of people in scene with the modal.The experimental results confirm that the proposed method have a high accuracy and robustness to the high-density crowd counting.