Moving Object Extraction Based on the Temporal Median Filtering for Image Intensity and Gradient Orientation Information

Moving object extraction is an important research area in computer vision with various applications, including gait analysis. The temporal Median filter (TMF) is a widely used technique for this purpose. The TMF can be applied to image intensities (INT) and also unit gradient vectors (UGVs). INT-based TMF and UGV-based TMF have different strengths and weaknesses. To combine these two techniques effectively, this paper proposes the automated thresholding method. The proposed method is based on Otsu’s thresholding followed by Normalized Cumulative Histogram (NCH) thresholding. We evaluate the integration framework and thresholding method on public datasets using F1-score and F2score as evaluation metrics. Our results show that the proposed fusion method works better than the INT-based alone or the UGV-based alone approach.