Toward a Flexible System for Pedestrian Data Collection with a Microsoft Kinect Motion-Sensing Device

Data on pedestrian activity, including volumes, walking speed, and trajectories, are used by transportation agencies and researchers for planning, design, and analysis. Several technologies are available for automatic collection of pedestrian data; however, all have inherent limitations in either functionality or monetary cost. Also, most technologies provide only counts. This paper proposes the use of an inexpensive motion-sensing device, Microsoft Kinect, which can track multiple people in low light conditions and can be combined with existing video-based daytime tracking. The tracking software and speed estimation methodologies are described, and indoor and outdoor studies show the system's effectiveness at determining pedestrian volumes and walking speeds. The accuracy of speed data is very satisfactory, with correlation of 98% or more for video data validation speeds. The accuracy of pedestrian volume data varies with traffic conditions; however, in low to moderate traffic conditions its performance is acceptable, with an undercounting error near 8%. The applications of the sensor and its complementarity with other sensors are discussed, as these are the first step toward a multisensor system.