Self-Adaptive Online Trajectory Sampling (SAOTS) Using Spectral Domain Properties

Abstract

Abstract We study the vehicle trajectory compression problem in this paper to reduce data size and transmission cost, while ensuring that the reduced data can satisfy the needs of transportation applications. We first propose an online segmentation method and then explore the spectral domain properties of vehicle trajectory segments by using the speed profile as the input signal. Based on the analysis, we introduce a spectral domain property called critical frequency that reflects the nature of the speed variation experienced by an individual vehicle trajectory. We show that, this property can be utilized to determine the proper sampling strategy of trajectory segments. We then design SAOTS, a self-adaptive online trajectory sampling method, for vehicle trajectory segmentation and compression. SAOTS utilizes vehicle flow state to segment a vehicle trajectory, and the critical frequency to sample the data records in the segment. Results show that resampling the trajectory segments based on the critical frequencies can vastly reduce the size of the GPS data while conserving most of the useful information. Tested using a small field dataset and a large, open dataset and compared with existing trajectory compression methods, SAOTS can help achieve the best balance in terms of data reduction, latency, accuracy, and application performances.