Ship trajectory data shows the characteristics of low precision, sparseness and trajectory drift for the port parking area recognition. To improve the accuracy of port parking area recognition based on ship trajectory big data, a Multi-constrained and Parallel Track Stay Segment Extraction (MPTSSE) method was proposed. Firstly, the definition of stay segment based on ship trajectory data was given as a basic concept for parking area identification. Secondly, a stay segment extraction model based on multiple constraints, such as speed, time difference, dwell time and distance, was introduced. Furthermore, a parallel trajectory stay segment extraction algorithm was proposed. Finally, Hadoop framework was adopted to implement the proposed algorithm. In comparison experiments with the trajectory stay segment extraction method based on Stop/Move model based on real ship trajectory big dataset, the accuracy of MPTSSE is increased by 22% in berth recognition of three ports. The MPTSSE method can effectively avoid misdivision of track stay segment and has better execution efficiency under large-scale ship trajectory dataset.

The traditional road information extraction method is high-cost and slow-update. Compared with it, road or marine lane information extraction from crowdsourcing trajectory data is low-cost and easier to update. However, it is difficult to extract lane boundary due to vessel trajectory data with high noise, large data volume and uneven distribution across different regions. To solve this problem, an extraction method of marine lane boundary from exploiting trajectory big data was proposed. Firstly, the parallelized denoising, interpolation and trajectory segmentation for trajectory big data was conducted. Then, based on parallelization and Geohash-encoded spatial clustering, trajectory data was simplified into multiple square regions. The regions were divided and the NiBlack method was extended as SpatialNiBlack algorithm to recognize regions on lane. Finally, based on the filtering results, del-alpha-shape algorithm was proposed to construct a Delaunay triangulation network and obtain marine lane boundary. The theoretical analysis and experimental results show that the proposed method can achieve an accuracy of 86.7% and a recall rate of 79.4% when the maximum density value is 200, minimum density value is 10, length and width of window are 5 and 5 respectively. The experimental results show that the proposed method is effective to extract valuable marine lane boundaries from large-scale trajectory data.

The discovery of travelling companions based on processing and analysis of the license plate recognition big data has become widely used in many aspects such as the involved vehicle tracking. However, discovery algorithms of travelling companions have poor performance in single machine mode no matter in time and space. To solve this problem, a discovery method of travelling companions named FP-DTC was proposed. This method based on the algorithm of FP-Growth was parallelled by the distributed processing framework-Spark, and had made some improvement and optimization to discover the travelling companions more efficiently. The experimental results show that, this method performs well on the discovery of travelling companions, and achieves an increase of nearly four times than the same algorithm with Hadoop.