University of Washington

PEDESTRIAN DETECTION USING MICROSOFT'S KINECT® Pedestrian movement data including volumes, walking speeds, and trajectories are essential in transportation engineering, planning, and research. While traditional image-based pedestrian detectors provide rich information, their performance degrades quickly with increased occurrence of occlusion, when one person is partially obscured by another. The 3D sensing capabilities of Microsoft's Kinect® present a potential cost-effective solution for occlusionrobust pedestrian detection. Kinect® was developed as a natural human interaction interface for the Xbox 360® game system. With its multi-sensor array, it provides both color video and depth information in real time, allowing occluded pedestrians to be separated and counted individually. This capability represents a significant improvement over conventional video-only approaches. The STAR Lab research team (Dr. Xiaofeng Chen, Mr. Kristian Henrickson, and Dr. Yinhai Wang) at the University of Washington developed a new algorithm to track and count pedestrians using the combined video and depth (RGBD) data from the Kinect® hardware. The algorithm was implemented using Microsoft Visual C# and OpenCV, and tested in three different crowded scenarios. It was proven effective with an average detection accuracy of 93.1%. The results demonstrated the feasibility of using the low cost Kinect® device and the proposed detection approach for real-world pedestrian detection in crowded scenes. Occluded pedestrians in a crowded and cluttered room are detected using the combined depth and color image from Kinect® Color image representing the depth information returned by Microsoft's Kinect® device Contact: Dr. Yinhai Wang Email: yinhai@uw.edu PACTRANS PROJECT SPOTLIGHT ON SAFE AND LIVABLE COMMUNITIES A UTC project led by Dr. Mike Dixon at the University of Idaho lives up to its multifaceted nature with a number of new developments on several fronts. Randal Brunello built the foundation for his grad thesis by completing a performance measurement test bed tool that facilitates exploring different applications of traffic controller data. This was recently demonstrated by a new method that leverages this data to estimate turning movement counts at intersections. UI student Stephen McDaniel created a new tool to spatially interpolate bicycle count data with promising results for a case study community. In practice, limited resources restrict count data collection to a few sites, as illustrated by the red dots on the map shown on the left. The best planning decisions enhance bike travel through a network. However, planners have difficulty translating these data points into bike volumes on routes between count locations or throughout the network. Stephen's tool uses a city's street network, land use, and any count data to interpolate, or estimate bike volumes on other portions of the network, for which no count data exist. Using this tool, practitioners can better anticipate bike network usage and network improvement impacts. Co-investigators David Porter and David Kim worked closely with ODOT to meet their need for inexpensive travel time data collection using Bluetooth technology. They expanded their system's capabilities and developed an economical system that anonymously monitors traveler movements, focusing on passenger vehicles. The new data collection system is the first of its kind to use Bluetooth data to track vehicles as they approach and leave a data collection point. This system will be a cost-effective means to anonymously measure urban street system performance, offering a variety of performance measures, in addition to travel time, such as delay. Even if the primary travel mode is not walking, ultimately, we begin and end all of our trips as pedestrians. Historically speaking, pedestrian data is difficult and expensive to collect. UW grad student Kristian Hendricks hopes to ease data collection with a new on-the-market technology found in the X-Box Kinect, that shows potential, even in difficult crowd conditions. He performed initial tests to assess Kinect's potential and will continue testing and development. This product aims to improve practitioner capabilities to quantify pedestrian flows and design more livable communities. Contact: Dr. Michael Dixon Email: mdixon@uidaho.edu Occluded pedestrians in a crowded and cluttered room are detected using the combined depth and color image from Kinect® 2012-2013 Annual Report 23

• Cover