PATH Draft Research Report Draft No . D 2006 - 14 Experimental Vehicle Platform for Pedestrian Detection

Pedestrian safety is a primary traffic issue in the urban environment. According to national traffic safety data, more than 4,000 fatalities were caused by pedestrian-vehicle collisions annually among the 40,000-plus total fatalities in all categories of collisions. Pedestrian crashes account for about 80,000 collisions per year, according to the General Estimate System (GES) data from 1995 to 1998. Over this time, there were, on average, 6000 pedestrian fatalities per year, accounting for about 15% of all roadway fatalities. Pedestrian deaths are primarily an urban problem. Many pedestrians are killed at crosswalks, sidewalks, median strips, and traffic islands. Seventy percent of pedestrian deaths in 2002 occurred in urban areas. According to GES data, 55% of all pedestrian crashes occur at non-junctions, while 40% occur at intersections or intersection-related locations. Similarly, in California statewide statistics, more than 3,500 fatalities resulted from 75,000+ pedestrian collisions in the last five years. A regional study has revealed a significant exposure to pedestrian accidents by transit agencies. The detection and tracking of pedestrians is not trivial. First, in an urban environment the patterns of vehicle and pedestrian movements can be very complicated. On city streets, there are numerous locations where pedestrians may choose to suddenly cross in front of a vehicle. The reliability of the detection function is further complicated by the crowded background and the issue of varying visibility, weather and roadway conditions. Thus, the solution for a reliable and accurate means of identifying pedestrians requires sophisticated design and extensive experimental evaluation. Given the challenging operating environment of transit buses, a complete and reliable sensing system for pedestrian detection can benefit from the combined use of multiple sensors. For example, the requirements of pedestrian detection are different for situations when buses are stopped at bus stops or near intersections versus when the buses are moving at relatively higher speeds in cruising conditions. Furthermore, any one particular technology may have difficulty meeting all necessary requirements under various lighting conditions, or rainy and foggy weather conditions, not to mention that most sensors have too limited a field of view to monitor traffic in all directions. In addition, the cluttered background and complex moving patterns of all objects on urban streets demand sophisticated processing of sensor inputs to avoid false detection and recognition. The work presented in this project represents a two-year effort taken by the research team to address pedestrian safety. With a survey of the latest …