In a conventional traffic lights controller, the lights either change at constant cycle times or at times proportional to the length of each leg of the intersection. Such approaches clearly are not perfect for optimizing traffic flow. Waiting times proportional to lane length may work well for a single-lane road but when roads with multiple lanes are considered the solution would not be optimal. The authors believe that an adaptive signaling based on fullness of each leg of the intersection would be a better approach. This paper presents the segmentation of foreground objects from frames of the surveillance video using an adaptive K-Gaussian mixture model and describes an approach for determining the lane occupancy rates for the north leg of the intersections. To give an accurate fullness measure the cast shadows that might be present in the segmented foregrounds are removed using a combined probability map called the shadow confidence score. Simulation results are provided for two standard and one custom recorded sequence.