RobustTP: End-to-End Trajectory Prediction for Heterogeneous Road-Agents in Dense Traffic with Noisy Sensor Inputs

We present RobustTP, an end-to-end algorithm for predicting future\ntrajectories of road-agents in dense traffic with noisy sensor input\ntrajectories obtained from RGB cameras (either static or moving) through a\ntracking algorithm. In this case, we consider noise as the deviation from the\nground truth trajectory. The amount of noise depends on the accuracy of the\ntracking algorithm. Our approach is designed for dense heterogeneous traffic,\nwhere the road agents corresponding to a mixture of buses, cars, scooters,\nbicycles, or pedestrians. RobustTP is an approach that first computes\ntrajectories using a combination of a non-linear motion model and a deep\nlearning-based instance segmentation algorithm. Next, these noisy trajectories\nare trained using an LSTM-CNN neural network architecture that models the\ninteractions between road-agents in dense and heterogeneous traffic. Our\ntrajectory prediction algorithm outperforms state-of-the-art methods for\nend-to-end trajectory prediction using sensor inputs. We achieve an improvement\nof upto 18% in average displacement error and an improvement ofup to 35.5% in\nfinal displacement error at the end of the prediction window (5 seconds) over\nthe next best method. All experiments were set up on an Nvidia TiTan Xp GPU.\nAdditionally, we release a software framework, TrackNPred. The framework\nconsists of implementations of state-of-the-art tracking and trajectory\nprediction methods and tools to benchmark and evaluate them on real-world dense\ntraffic datasets.\n