moving shadow and object detection in traffic scenes.pdfVIP

Appears in: 4ICPR 2000 Moving Shadow and Object Detection in Traffic Scenes Ivana Mikić, Pamela C. Cosman, Greg T. Kogut, Mohan M. Trivedi Department of Electrical and Computer Engineering, University of California, San Diego, USA Abstract contour of the shadow. However, this is rarely true for outdoor scenes, where shadows usually have sharp edges We present an algorithm for segmentation of traffic and background is often non-textured. Even with scenes that distinguishes moving objects from cast textured background, our experience is that texture is shadows. Three image features at each pixel site are almost invisible in the shadow regions due to the considered: brightness, and normalized red and blue properties of the imaging process. color components. Each feature is analyzed by an a posteriori probability estimator that computes Without using scene models and assumptions mentioned membership probabilities for the three classes: above, we can identify three sources of information that background, foreground and shadow. The algorithm can help in detecting objects and shadows. The first is iterates between the three estimators by using the output local, based on the appearance of the individual pixels. A of one as the a priori probability input to the n