Map Matching for Uncertain-Coordinate-Encryption GPS Trajectories Based on Light-Weight Convolutional Neural Network

Map-matching is an important technique usually conducted by associating a series of recorded geographic coordinates into a route on the road network to make GPS trajectory data available for the vehicle traveling route inference. In practice, some GPS companies apply multiple encryption processes into their GPS points to satisfy the national safety requirement. This paper proposes an image recognition based map matching (IR-MM) system that can infer accurate vehicle traveling routes for uncertain-encryption trajectories from the regulatory platform. In the proposed system, a light-weight convolutional neural network (CNN) model is introduced to extract feature vectors of trajectories image and road routes image that reduce the utilization of the geographic distance in route inference, unlike the very most map-matching algorithms. Furthermore, we also enhance the algorithm performance by performing trajectory clustering and K shortest routes searching. We evaluated the performance of our system using vehicle driving datasets including hundreds of real trajectories with low sampling rate. The experimental results demonstrate that our system has higher matching accuracy (91.2%) in comparison with four existing methods and its average execution time per trajectory (2.5 min) is comparable to these four algorithms.