Crowd motion estimation and motionless detection in subwaycorridors by image processing

摘要

In this work, we tried to provide, in the context of crowdmonitoring in subway (underground railway) station corridors, a motionestimation method and a motionless detection method which both take intoaccount three difficulties: real time constraint, deformable objects andoccultation. The first part of our study concerns motion estimation.Thereby, we chose three motion estimation techniques, each of thembelonging to one of the three main classes of motion estimation methods.The chosen techniques are: block matching (primitive matchingtechnique), optical flow (differential technique) and Gabor filter(frequential technique). We then attempted to improve both the speed andthe performances of each algorithm. As a result, we got a very efficientoptical flow technique suitable for various applications in the field ofcrowd monitoring and traffic management. The second part of our studyconcerns the motionless detection. Thus, we developed a method based onmotion detection and background image construction