Representative pedestrian collision injury risk distributions for a dense-urban US ODD using naturalistic dash camera data

Automated Driving Systems (ADS; SAE levels 3 through 5 technologies) are currently being deployed in several dense-urban operational design domains (ODDs) within the United States (US). Within these dense-urban areas, vulnerable road users (VRU) generally comprise the vast majority of injury and fatal collisions. One challenge with the study of VRU collisions is a lack of crash data sources with pre-impact kinematics. Understanding the pre-impact kinematics is a key factor in assessing the potential injury risk for pedestrian-vehicle impacts. The purpose of this study was to determine injury distributions for pedestrians within a dense-urban ODD (Los Angeles, California) using data from vehicles instrumented with forward-facing cameras and vehicle sensors. This study leveraged data from a fleet of vehicles equipped with aftermarket, in-cabin dash cameras operating in Los Angeles, California. From approximately 66 million miles of driving data, 42 collisions were identified. Each vehicle was equipped with a forward-facing camera, an accelerometer sampling at 20 Hz, and GPS. A global optimization routine was used on the accelerometer, GPS, and video data to correct for sensor orientation and asynchronicity in data sampling. For each event, two key video frames were identified: the frame associated with impact and a frame associated with key vehicle kinematics (e.g., vehicle start/stop, hard braking [> 0.2 g]). These key frames were then mapped to the processed vehicle speed kinematics to determine vehicle speed at impact. For the events included in this dataset, impact speeds ranged from approximately 1.6 kph (1 mph) to 65 kph (40 mph). In most events, the front of the vehicle struck the pedestrian. Existing pedestrian injury risk curves were then used to calculate the level of risk associated with the reconstructed impacts, and the probability of AIS3+ injury risk was observed to vary from minimal risk (<2%) to approximately 55%. These data highlight the wide range of impact speeds and injury risk that may occur during vehicle-pedestrian collisions. Assessing injury severity for collisions involving VRUs is highly impactful for the continued development of traffic safety, including ADAS, ADS, and roadway design. Using naturalistic VRU collision data collected from dashboard cameras, a methodology for assessing event severity by pairing accelerometer and GPS data with video to compute impact speed was presented. This is the first known analysis of pedestrian severity distributions using a naturalistic US database. The methods presented in this study may be applied to larger datasets or other sensing systems to enable further ODD-specific modeling.