Real-time Detection Of Abnormal Ship Trajectory Based On Trajectory Clustering And KDE

The growth of the water transportation industry,increased traffic density,and the development of large-scale and specialized ships have created a more complex navigation environment,increasing the difficulty of supervision for the supervisory departments of the port while also increasing the risk of ship navigation.Anomaly ship trajectory detection is important for the safety of ship navigation.Its goal is to increase the automation and intelligence of maritime supervision in order to reduce the number of water accidents and illegal acts,improve traffic efficiency,and ensure maritime navigation safety.Therefore,this dissertation proposes a method for detection of anomaly ship trajectories based on trajectory clustering and KDE(Kernel Density Estimation).This paper is mainly divided into three parts:ship trajectory segmentation,ship motion pattern extraction and the application of real-time detection of abnormal ship trajectory.1)Ship trajectory segmentation based on ADP algorithmThe method proposed in this paper includes ship trajectory segmentation,which is an important part and one of the key technologies.The most commonly used method in trajectory segmentation is the compression algorithm,and most studies on ship trajectory compression have one or more of the following shortcomings:low compression efficiency;error ship static information when the distance threshold is set based on ship length or width;and difficulty in setting compression threshold.To solve these problems,this paper proposes the ADP(Adaptive-threshold Douglas-Peucker)algorithm based on the improved DP(Douglas-Peucker)algorithm.The ADP algorithm determines the threshold value of each trajectory through the threshold change rate of the trajectory points,which does not rely on the static information of the ship,and is easier to determine the threshold value.In addition,the ADP algorithm uses matrix operations and the method of reducing trajectory points to improve the computational efficiency.The experimental results show that,when compared to the commonly used DP algorithm,the improved DP algorithm and the Sliding Window algorithm,the ADP algorithm can maintain a higher similarity to the original trajectory at the same compression rate,and its threshold is easier to set,and the compression effect is unaffected by the ship’s static information error and the size of the water area.2)Ship motion pattern extraction based on novel clustering methodThis paper proposes a novel trajectory clustering method for extracting ship motion patterns.Trajectory clustering is the main method for ship motion pattern extraction based on AIS data.Previous research on the methods of trajectory clustering is mainly divided into three categories:overall trajectory clustering,sub-trajectory clustering and trajectory point clustering.The first method tends to ignore most of the local features,and the longer the trajectory is,the more features will be ignored;the second tends to ignore the overall characteristics of the ship’s trajectory and cannot identify the overall movement trend of the ship;the third is often used as auxiliary tools.Clustering algorithm used for trajectory clustering is very important,and different algorithms may lead to completely different results.There are issues with the four commonly used clustering algorithms,K-means,K-medoids,spectral clustering and DBSCAN(DensityBased Spatial Clustering of Applications with Noise),such as parameter setting difficulty,noise recognition ability,and sensitivity to the density distribution of the dataset.In addition,clustering algorithms are also exclusive,that is,an object belongs to only one cluster.In order to address these issues,this study proposes a ship trajectory clustering method(FOLFST)that can find the overall and local features of ship trajectories.FOLFST clustering method considers the similarity between the sub-trajectories and the similarity between the mother trajectories when measuring the similarity of the sub-trajectories,which can well identify the overall and local features of the ship’s trajectory;and FOP-OPTICS algorithm is used to cluster the sub-trajectories.Finally,FOLFST clustering method identifies the sub-trajectories belonging to multi-cluster class to solve the problem of exclusivity of clustering.FOP-OPTICS clustering algorithm adopted by FOLFST is an improved clustering algorithm based on OPTICS,which finds the demarcation point from the Augmented Cluster-Ordering generated by OPTICS and uses the reachabilitydistance of the demarcation point as the radius of eps-neighborhood of its corresponding cluster.It overcomes the weakness of most algorithms in clustering datasets with uneven densities.By computing the distance of the k-nearest neighbor of each point,it reduces the time complexity of OPTICS;by calculating density-mutation points within the clusters,it can efficiently recognize noise.The experimental test shows that the accuracy,sensitivity,specificity and AUC of FOLFST clustering results are 94.96%,97.17%,80.33%and 88.75%,respectively,which are much higher than test results of the other several algorithms.3)Real-time detection of abnormal ship trajectory based on FOLFST-KDECombined with the KDE(Kernel Density Estimation,KDE)method,this paper proposes a real-time detection method for abnormal ship trajectory based on FOLFST-KDE.FOLFST-KDE uses the FOLFST clustering method to extract the motion pattern of the ship,and then combines the KDE method to calculate the representative position distribution(RPD)and the representative speed and course distribution(RSCD)of the ship trajectory.RPD is used to detect the abnormal situation of ship position in real-time,and RSCD is used to detect the abnormal situation of ship speed and heading in real-time.In order to verify the effectiveness and feasibility of the research method,this study selects the AIS data in the Qiongzhou Strait waters from November 01,2019 to November 21,2019 for experiments,and selects five sets of real ship data for the experiment(the five sets of data were crossing the channel,sailing in the reverse direction,part of the trajectory deviating from the channel,overspeed,and the ship’s trajectory in normal sailing),and the LSTM neural network commonly used in trajectory anomaly detection is selected as a comparison algorithm.The experimental results show that the proposed method can identify abnormal ship trajectories more quickly and accurately,and the detected abnormal results are more interpretable.