Toward Opportunistic Compression and Transmission for Private Car Trajectory Data Collection
Jie Chen, Zhu Qiang Xiao, Dong Wang, Daiwu Chen, Vincent Havyarimana, Jing Bai, Hongyang Chen · IEEE Sensors Journal · 2018
The advances in vehicle location service and communication techniques have generated massive spatial-temporal trajectory data, which has caused the crises of storage and communication in the vehicle trajectory data center. In this paper, we propose a novel opportunistic compression and transmission-based long short-term memory method, namely, OCT-LSTM, with aims of reducing trajectory transmission overhead and storage cost. We first present a low-cost vehicle location device for trajectory real-time collection of private cars. Within the proposed OCT-LSTM, we introduced a map-matching method based on MIV-matching which reduces sampling errors of raw trajectories. Then, we present a spatial-temporal transformation method to divide the trajectory data into two parts, i.e., spatial path and time-distance parts, and realize the compression operation separately. Similar movement patterns are repeated and randomly present in trajectories of private cars. In this paper, we train the LSTM model to remember and predict these repetitive movement patterns through historical trajectories. An opportunistic transmission of trajectory data from the vehicle terminal to the data center was designed, which can dramatically decrease the transmission overhead. The proposed OCT-LSTM not only realizes real-time trajectory preprocessing and compressing but also ensures high trajectory compression ratio. To validate the performance of the OCT-LSTM, we collect a large-scale private car trajectory data from real urban environments. The experiments verify the compression ratio effectiveness and time-delay superiority of the proposed methods.