Pedestrian friendly traffic signal control : final research report.

Final research reportPDFTech ReportDTRT12GUTG11CamerasDetection and identification systemsPedestrian countsPedestrian flowPedestrian trafficTraffic signal control systemsUrban areasUnited StatesCarnegie-Mellon University. Technologies for Safe and Efficient Transportation University Transportation CenterSmith, Stephen F.Barlow, Gregory J.Hu, Hsu-ChiehHua, Ju-HsuanCarnegie-Mellon University. Technologies for Safe and Efficient Transportation University Transportation CenterUniversity Transportation Centers Program (U.S.)United States. Dept. of Transportation. Research and Innovative Technology AdministrationUS Transportation CollectionThis project continues research aimed at real-time detection and use of pedestriantraffic flow information to enhance adaptive traffic signal control in urban areaswhere pedestrian traffic is substantial and must be given appropriate attention andpriority. Our recent work with Surtrac [12], a real-time adaptive signal controlsystem for urban grid networks, has resulted in an extended intersection schedulingprocedure that integrates sensed pedestrians and vehicles into aggregate multi-modaltraffic flows and allocates green time on this integrated basis [17]. In this project weconsider the companion problem of providing the pedestrian sensing capabilitynecessary for effective use of this extended intersection scheduling procedure.Although some commercial pedestrian detection and counting capabilities do exist,they typically require the purchase and installation of additional higher resolutionvideo camera technology, which can double the cost of detection per intersection.Our interest is in a solution that does not significantly increase infrastructure cost.The hypothesis investigated in this work is that lower resolution vehicle detectioncamera technology can be used to provide a relaxed form of pedestrian count datathat is sufficient for incorporating pedestrian flow information into real-timeintersection scheduling. Specifically, we study the possibility of extracting anapproximate but usable measure of pedestrian \ue2\u20ac\u153density\ue2\u20ac? from the video stream of acommercial traffic camera. Our target functionality is the ability to qualitativelydiscriminate between \ue2\u20ac\u153no\ue2\u20ac?, \ue2\u20ac\u153few\ue2\u20ac? or \ue2\u20ac\u153many\ue2\u20ac? waiting pedestrians. Contemporarytraffic camera technologies provide resolution as low as 320 \uc3\u2014 240 gray scale images(see Figure 1), together with the ability to specify and monitor a set of occupancyzones within the image.Pedestrian detection and counting is not a hard task for humans, but it is challengingfor computers. The challenges include diverse shapes and occlusion amongpedestrians, a dynamic background, and low video quality. First, pedestrians canhave various appearances because of clothing, accessories, assistive devices, andchange of pose while walking. This high intra-class variation, as well as occlusion,makes pedestrian detection a hard problem. An alternative to classification basedpedestrian detection is to find foreground pixels in each frame of the video, andanalyze those pixels to estimate the number of pedestrians. However, since thesystem is deployed in an outdoor environment, shadows caused by moving objects orsudden change in illumination can create noise that complicates foregrounddetection. Moreover, to get a broader view of the intersection, the camera is installedat a certain height. Thus, pedestrians are small in the images and have fewer detailsfor computer vision processing