Motion estimation of crowd flow using optical flow techniques: A review

Abstract

Velocity and direction estimation plays an important role in crowd analytic and behavior recognition. This paper presents an overview of the literature published for motion detection and estimation techniques. The work particularly focuses on optical flow techniques such as Lucas & Kanade and Horn & Schunck methods which describe the direction and velocity of pixels in a sequence of two consecutive images. A two-dimensional velocity vector, carrying information of the direction and the velocity of motion is assigned to each pixel in a given place in the image. Optical flow method is extensively used for motion estimation due to its ability to compute the velocities accurately. The improvement in computational efficiency and increasing interest in robust and accurate motion estimation algorithms lead to increase in the use of optical flow in crowd analytic applications. We investigate the implementation of optical flow methods in the published work and we provide comparison between these techniques qualitatively as well as quantitatively. The qualitative analysis illustrates the optical flow performance in terms of rigid motion, non-rigid motion, motion discontinuities, noise and different light condition. Quantitative analysis is in terms of computational time and accuracy.