Feature extraction for moving object detection in a non-stationary background

Classification of moving objects in a non-stationary background has become a vast area of study, which is used in various applications where neither the background nor the foreground is stationary. Feature detection becomes an important part to detect moving objects when the background itself is also moving. Harris corner detection, SIFT (Scale Invariant Feature Transform) and SURF (Speeded-Up Robust Features) are three most commonly used feature detection algorithms. Clustering of feature points using block-based k-means clustering along with feature detection are used to distinguish moving objects from the background. This paper presents a study among various methods that were evaluated and compared on robustness, amount of points detected, computation time and overall performance. The simulations were carried out in MATLAB.