SaveDat: Spatio-Temporal Trajectory Compression by LSTM

The emergence of automatic navigation systems brought an exponential increase in location data transmitted by vehicles to cloud services, which opens the door to optimizing storage and reducing communication of this data. In addition, location data and trajectories are often overlapping in many cases, such as navigation, fleet management, and logistics. Utilizing the collected data from many vehicles can contribute for location data compression sent from the vehicle to the service, but such data is held by the service side and transmitting it to the vehicle is inefficient in bandwidth and storage. As such, it would be advantageous if the vehicle could be made aware of common patterns of location-based trajectories in order to efficiently compress its location transmissions to the service or even mute while the expected trajectory keeps an acceptable error. We propose SaveDat - a Spatio-temporal Trajectory compression solution that discerns route patterns from location data at the service side and shares those patterns with vehicles in order to fit actual trajectories without sending location data or compressing it.Experiments present a compression ratio of up to 100:1 in typical scenarios.