In this article we present a novel algorithm for detecting moving objects using a stationary camera, which is based on statistical background modeling using edge segments. Traditional algorithms that rely on pixel intensity struggle in dynamic environments due to their inability to handle sudden changes in lighting conditions. These algorithms also tend to produce ghosting artifacts when a sudden change occurs in the scene. To address this issue, edge-based features that are robust to intensity variations and noise have been introduced. However, existing methods that rely on individual edge pixels suffer from scattered edge pixels and cannot effectively utilize shape information. Additionally, traditional segment-based methods struggle with variations in edge shape and may miss moving edges that are close to the background edges. In contrast to conventional approaches, our proposed method constructs the background model using regular training frames that may include moving objects. Furthermore, it avoids the generation of ghosting artifacts. Additionally, our method employs an automatic adaptive threshold for each background edge distribution to facilitate matching. This enhances the robustness of our approach to changes in illumination, camera movement, and background motion. Experimental results demonstrate that our method outperforms other techniques and efficiently detects moving edges despite the aforementioned challenges.
Keywords: motion detection, edges, canny edge detector, gaussian of color, gaussian of gradient magnitude, normal distribution, adaptive thresholds, statistical map