Abstract
In Japan, the pedestrian occupies the most fatalities of traffic accidents. The characteristic of a pedestrian fatal accident is that the majority of cases occur during crossing. In the conventional pedestrian detection, only the pedestrian in front of the own car is detected. Therefore, the detection is suddenly and the reaction is difficult. This paper proposes the detection method which has a margin of time by detecting intention and danger of crossing before a pedestrian cross the roadway. Proposed method adopts Deep Neural Network as the machine learning method used for pedestrian detection, and the results of the different input data form are compared. This paper shows the details of the proposed method and the results of application to videos of the drive recorder.
