Detection of the human skeleton on pedestrians in road traffic

Abstract:
Since pedestrians are vulnerable road users compared to vehicles, many of them are still injured and killed in road accidents. The reason for this is either misconduct of the pedestrians themselves or that the drivers overlook them. To realize pedestrian protection, detection and warning systems need accurate information about the current as well as future position and action of pedestrians. Automated vehicles must be able to predict the paths and intentions of pedestrians to generate a warning or adjust their path, if necessary, to avoid a collision. These aspects can only be handled if a large number of exteroceptive sensors is used to capture the environment with the greatest possible level of detail and detect objects such as people. The estimation of joints and body parts can then provide information about the movement of a pedestrian and thus be further used to determine current and future activity. Previously known methods for tracking the human skeleton seem to have partial problems with the detection of pedestrians in the environment of automated driving. For example, road signs or other road traffic objects are detected and estimated as a skeleton. This weakness is to be counteracted with independent preliminary pedestrian detection.