영문내용
(English Abstract) |
Moving object detection has become a very emerging research area because of its broad range of applications. Some manual image processing- and deep learning-based methods have gained success in the moving object detection for the non-dynamic video. However, the approaches fail to detect moving objects for the dynamic video case such as dynamic background, unstable video, camera jitter, pan-tilt-zoom, and bad weather. To address this problem, we propose a background subtraction-based moving object detection method, called ODD. In this method, we use the only color feature of a pixel for the effective segmentation. The pixel features like the gradient magnitude and local binary pattern we do not use for this segmentation as the features introduce more extraneous noises for the dynamic video. We take a sample consensus-based policy for per-pixel segmentation. Based on background region behavior, we determine a per-pixel decision threshold. Finally, we do adaptive post-processing to discard blinking pixels. We consider the F-measure metric for the CDNet-2014 dataset to evaluate the performance of our approach. Our proposed approach outperforms for a pan-tilt-zoom video case. |
