Research On People-Flow Detection And Analysis From Video Sequences

With the development of modern socio-economic and the accelerated process ofurbanization, the density of population grows quickly. It could be over-crowded in thepublic areas which increase the risk of accidents. So people monitoring and analysispublic such as stores and stations is of great significance. People flow detectionand analysis system based on video can implement continuous monitoring, analysisand alarm for a people area, which has great significance for reality and applicationprospect, and it is also an important direction of current research.The algorithm of people flow detection and analysis from video sequences can bedivided into three main parts: people flow detection algorithm, people flow trackingalgorithm and people flow counting algorithm. This paper mainly discusses these threeparts, proposes and implements a framework of people detection and analysis fromvideo sequences, which detecting, tracking and analyses people flow, and makes analert when people density reach a certain threshold.The objective of people flow detection is extracting the people flow area and individualsfrom the video sequences. In this paper people flow area is extracted using acombination of the running average-like background subtraction and the edge imagesdetection, which is much less sensitive to lighting condition changes and more efficientand robust. Individual detection faced occlusion and segmentation problem. Theocclusion problem can be solved effectively with an overhead mounted camera. Thesegmentation problem can be solved using k-means clustering with people number inthe area as a priori, which can be estimated by a linear relationship between foregroundpixels and the number of people.The objective of people flow tracking is establishing corresponding relationship with individuals in the continuous video sequences, tracking individuals in the peopleflow, researching on the behavior of people flow and individuals in-depth. This papercompares several popular object tracking algorithms. Then according to some characteristicsof the people flow tracking, a tracking method based on motion predictionhas been adopted, , which is more efficient and robust, and can meet the needs of thereal-time people flow tracking.The objective of people flow counting is to count the number of pedestrians througha region for a period of time. This paper designs an effective people flow counting algorithm,which can estimate people flow density and make the judgment about the safetyof the region.This paper provides important reference for the future research on the real-timepeople flow detection and analysis. People flow detection and analysis system will beapplied in mobile devices in the future. People flow detection and analysis, includingreal-time video monitoring for a people area, researching on the behavior of peoplein-depth and making an alert when people density reach a certain threshold, is of greatsignificance to the safety and the development of the society.