It is both commercially beneficial and safety concerning to count the number of pedestrians. The owner of commercial real estate can adjust strategies to improve the sales according to counting results. The safety manager can monitor the flow of crowds and maintain public safety. Of all the methods applied, vision based pedestrian counting methods gain popularity because of high accuracy and little physical contact. However, these methods are subject to environmental factors such as illumination changes, shadows, occlusions and similarity in background. In this case, the performance of vision based pedestrian counting methods is degraded.This paper proposes a different pedestrian counting method using a laser range scanner that addresses these problems effectively. Height features of pedestrians are used to analyse the height map, which help to improve the counting accuracy. The background is built from data retrieved from laser range scanner when no pedestrians appear in experimental sites. Foreground height is extracted from newly retrieved data and background. Height map is formed by accumulating foreground height in a time interval. A scanning point clustering algorithm is proposed to extract height clusters from height map. Afterwards, Gaussian regression process is applied to count the number of pedestrians in each height cluster. A variable-sized slide window is used to detect heads in height cluster. The voting results of scanning points are gathered to determine the walking direction of that pedestrian. Two experimental sites are built to test the system. The experimental results in both indoor and outdoor sites prove the speed and accuracy of the system. The system is able to count the number of pedestrians in crowded scenes accurately. |