Accurate pedestrian counting is presented in scenarios where a moving background is encountered. It was determined through detailed observation that the human head presents one of the most invariant features of persons dressed normally and being observed, from above, some meters away in potentially crowded situations. A model-specified directional filter (MDF) is used to detect candidate object locations and a simple matching process to identify the pedestrian heads through a temporal sequence of acquired images. The method fails to find the initial search points only if the contrast of the image is too low. The matching process is not successful if only part of the pedestrian head edge can be detected as a result of a poorly contrasting image.
|Publication status||Published - Jul 1997|
|Event||6th International Conference on Image Processing and its Applications - Dublin, Eire|
Duration: 1 Jul 1997 → …
|Conference||6th International Conference on Image Processing and its Applications|
|Period||1/07/97 → …|