Suppression of shadows for improved object discrimination

The ESPRIT PEDMON (PEDestrian MONitoring) project is focusing on the development of technologies to enable the automated counting of pedestrians in unconstrained areas. A number of mechanisms are being investigated within the project, and image processing is taking a leading role in the development of wide area monitoring and counting systems. Initially, a simple approach to the video processing has been adopted, where the camera is placed vertically above the counting region allowing a 2D approximation of the scene to be made. In this approach the early part of the image processing pipeline consists of a segmentation procedure to separate the pedestrians from the background. A basic philosophy has been adopted that, for the purposes of segmentation, the pedestrians constitute moving areas of the image and anything not moving is considered to be background. This allows the use of well-established approaches for segmentation. We develop a mechanism which compares the texture of a reference frame with the texture of the incoming image to determine whether a shadow is present and subsequently attempt to suppress it. We have shown that shadows can be detected and suppressed in a number of scenes, providing that the background is already known