Pedestrian Orientation Estimation based on Super Resolution of LiDAR Data

Pedestrian detection for autonomous driving has been intensively studied in the past two decades. Especially, with the development of deep neural networks, the performance of pedestrian detection has been significantly improved. However, more details information is needed for autonomous driving in urban environments. Pedestrian body orientation is meaningful information to indicate the pedestrian's walking direction, and pedestrian orientation estimation from the camera sensor has been proposed in previous research. Still, there are few discussions for the pedestrian orientation estimation from LiDAR (Light Detection and Ranging) sensor. This research proposes to estimate pedestrian orientation from LiDAR data using deep neural network-based super-resolution to overcome the low-resolution issue of LiDAR sensors. The experimental results indicate that the proposed method can achieve satisfactory performance. The proposed system can be used as compensation for the camera-based pedestrian orientation estimation.