Foreground Detection based on Co-occurrence Background Model with Hypothesis on Degradation Modification in Background Changes
Wenjun Zhou, Shunrichi Kaneko, Manabu Hashimoto, Yutaka Satoh, Dong Liang · 2018
This paper presents a robust background model for foreground detection in background changes, such as illumination changes and background motions. By utilizing the co-occurrence “pixel to block” structure, our model, which is named as Co-occurrence Pixel-Block Pairs (CPB) can learn the change information of the background and extract the spatial-temporal information of each pixel in the scene, and build a prospective background model for each pixel. Then, we employ an efficient evaluation strategy to identify the current state of each pixel to classify the foreground and background, which is named as correlation dependent decision function. A key contribution of this work is it offers a robust background subtraction against the dynamically changing background. Furthermore, we also introduce a Hypothesis on Degradation Modification (HoD) into CPB structure to reinforce the robustness of CPB, and it also further improves the performance of our algorithm. This observation is robust to illumination changes and background motions. Experimental results obtained from the datasets under different challenges prove that our algorithm has a good performance for foreground detection.