At the Traffic Intersection, Stopping, or Walking? Pedestrian Path Prediction Based on KPOF-GPDM for Driving Assistance

Figure 1

KPOF-Net is composed of an optical flow module and a joint point module. On the one hand, the optical flow extraction of pedestrians and obstructions is completed through self-flow, WEC is proposed to remove obstructions, and then the UCTGAN network is used to generate a complete pedestrian optical flow diagram. On the other hand, it is based on B-GPDM training four types of joint models of walking, stopping, walking tendency, and stopping tendency. Finally, the pedestrian optical flow information is combined with the joint point model to predict the pedestrian’s posture and trajectory.