ONF-TRS: On-line Noise Filtering Algorithm for Trajectory Segmentation Based on MDL Threshold

摘要

Background: Spatial trajectories suffer from noise that may be caused by poor signal of GPS devices, sometime the noise is acceptable few meters from its true location. In different situations, the noise is too big that dramatically change the information derive from trajectory segments such as speed, thus filtering of noise is needed before starting mining task. Materials and Methods: The proposed algorithm on-line noise filtering for trajectories segmentation ONF-TRS segments trajectory points to set of significant points after removing non-significant and noise points. The key idea is both non-significant and noise points have small value of (region/length), which mean travel long distance and cover small region. The threshold value of (region/length) is estimated using minimum description length concept. Results: Experimental results in real data sets confirm the effectiveness of (ONF-TRS) algorithm in filtering noise points during segmentation process, while existing algorithms need to implement noise filtering step before segmentation. Conclusion: This study provides ONF-TRS algorithm appropriate for trajectories segmentation and spatial noise filtering simultaneously which makes the algorithm convenient for stream data mining.