The Study of Efficacy of Gaussian Mixture Model In Image Tracking System Using the Canny Optical Flow Technique

Abstract

Detecting and tracking moving objects in a sequence of video images is an important operation in the field of computer vision. However, problems of image noise, complex object forms and motions, and real-time processing are few of the challenging facing existing flow calculation methods as they are computationally complicated and highly vulnerable to noise. Hence, this paper presents a technique less susceptible to noise and computationally efficient. Canny Optical Flow (Canny-O-Flow) was used in this study to develop a framework for recognizing and tracking moving objects in video. Video datasets in avi and mp4 format were acquired from MathWorks and YouTube respectively. The obtained datasets were preprocessed by splitting the videos into numerous frames. The video dataset's rapid shift in intensity and Object's borders were discovered using Gaussian Mixture Model and Canny Edge technique, respectively. The performance of Canny-O -Flow and O-Flow techniques were investigated by compared using Accuracy, Precision, False Acceptance Rate, False Rejection Rate and Processing time as the performance metrices. Additionally, the techniques were validated by t-test to compare the differences between the two techniques at 5% significance level. It showed that p=0.008 for test of significance between the Canny-O-Flow and O-Flow techniques. The developed Canny-O-Flow framework was able to recognize and track moving objects more efficiently and with less processing time than the standard technique.