Egocentric Two-frame Pedestrian Trajectory Prediction Algorithm Based on a Panoramic Camera

Abstract

With the development of service industry, service-oriented mobile robot navigation technology has received more and more attention. In order to improve the obstacle avoidance performance of service-oriented mobile robots, accurate pedestrian trajectory prediction is essential. Therefore, we propose an egocentric two-frame pedestrian trajectory prediction algorithm based on a panoramic camera, the Future Social Prediction (FSP) algorithm, to solve the pedestrian trajectory prediction problem in the human-robot coexistence environment. The input of FSP is the information in the two frames of a panoramic camera. First, the free pedestrian prediction network (FPPN) is used to predict the free movement of pedestrians. Then the future social pooling module (FSPM) is used to process the prediction results of FPPN. And finally, the social pedestrian prediction network (SPPN) combines the frame information with the output of FSPM to get the final prediction results. A panoramic camera is used as a sensor to reduce blind spots in the field of view. Only two frames are used for prediction to improve the real-time performance of FSP. We record the panoramic video and create a dataset. FSP is tested on our dataset with other two algorithms, and the results show that FSP has better performance than Spatio-Temporal Encoder-Decoder model (STED) and Future Person Localization (FPL).