Real-time detection of crossing pedestrians for traffic-adaptive signal control

We propose an approach to detect and count pedestrians at an intersection, in real-time, using a fixed camera. After identifying moving objects in sequential images via motion segmentation, median filtering and erosion/dilation operations are performed to suppress noise. Connected component extraction is then employed to extract and label the moving objects in the image. The binary foreground region of each component is projected onto the axis perpendicular to its major axis to obtain a vertical projection histogram from which shadows can be detected, extracted and suppressed. Information about the size and coordinates of each component is then utilized to compute the number of people in the scene.