Deep neural architecture for geospatial trajectory completion over occupancy gridmap

GPS data is widely used in many real world applications. The quality of GPS data is critically important to produce high quality results. In real world applications, certain GPS trajectories are sparse and incomplete; which causes challenges to GPS trajectory based applications. Few of existing studies have tried to address this problem using complicated algorithms that are based on conventional heuristics; this requires extensive domain knowledge of underlying applications. Deep learning in recent era has achieved the great success to solve many sequence to sequence prediction problems. In this paper, a deep learning based bidirectional convolutional recurrent encoder decoder architecture using attention mechanism is proposed that predicts the missing data points; resulting in a complete GPS trajectory. The proposed method shows the significant improvement over state-of-the-art benchmark methods.