People are counted in a segment of video with a video processing system that is configured with a first set of parameters. This produces a first output. Based on this first output, a second set of parameters is chosen. People are then counted in the segment of video using the second set of parameters. This produces a second output. People are counted with a video played forward. People are counted with a video played backwards. The results of these two counts are reconciled to produce a more accurate people count.