Frame differencing, a single Gaussian, and modified GMM for foreground object detection on camera jitter movies in comparison to F-Score measurement

Abstract

Aim: The aim of the study is to evaluate the performance of the novel Modified Gaussian mixture model algorithm in detecting foreground objects on camera jitter videos by comparing it with frame differencing, Single Gausian method, Gausian mixture model in four different scenarios. Materials and methods: A total of 6420 frames were taken from Badminton, Boulevard, Sidewalk, Traffic datasets and those video sequences have 1150, 2500, 1200 and 1570 frames respectively.The dataset involves two groups based on indoor and outdoor environments. The indoor group involves 1150 frames and the outdoor group consists of 5270 frames. Precision, Recall and F-Score were calculated to evaluate the performance of all the algorithms. Results: The mean F-Score values of frame differencing, single Gaussian, Gaussian Mixture Model and Modified GMM model algorithms are 0.1874, 0.0149, 0.3030 and 0.2927 respectively. The Modified GMM model and GMM model provided better average F-Score and it is significantly better than that of remaining two models (p<0.0001, =0.05, power=80%). Conclusion: From this study it is observed that the novel Modified GMM and GMM algorithm performed better than the other two algorithms in these four video sequences.