Trajectory Outlier Detection on Trajectory Data Streams

Abstract

The detection of abnormal moving on trajectory data streams is an important task in spatio-temporal data mining. An outlier trajectory is a trajectory grossly different from others, meaning there are few or even no trajectories following a similar route. In this paper, we propose a lightweight method to measure the outlier in trajectory data streams. Furthermore, we propose a basic algorithm (Trajectory Outlier Detection on trajectory data Streams-TODS), which can quickly determine the nature of the trajectory. Finally, we propose an Approximate algorithm (ATODS) to reduce the detection cost. It is space approximate algorithm which can effectively reduce the amount of calculation. The cost of ATODS algorithm can satisfy the demand of trajectory data streams. Our method are verified using both real data and synthetic data. The results show that they are able to reduce the running time without reducing the accuracy. INDEX TERMS Outlier, trajectory stream, moving object.

Cleaning Big Data Streams: A Systematic Literature Review

“…The tropical atmosphere ocean dataset and the generated dataset were utilized for the experiment, and the results were compared with those achieved using the incremental local outlier factor algorithm in terms of time requirements, space requirements, and F-score. Cao et al [77] suggested a lightweight method and approximate trajectory outlier detection using the trajectory data stream algorithm to detect global outliers, and the proposed method was compared to trajectory outlier detection using trajectory data streams and isolation-based anomalous trajectory algorithms in terms of running time and accuracy. Dias et al [78] achieved a decreased anomaly score by using a repeated sequence algorithm to detect global outliers.…”

Outlier Vehicle Trajectory Detection Using Deep Autoencoders in Santiago, Chile

“…For the identification of outliers they use the reconstruction error so that, if the time series has a high reconstruction error, then it is considered an outlier. Cao et al [ 23 ] introduce a neighborhood-based time stream outlier detection algorithm. This method is based on the estimation of the neighborhood of trajectories in a database considering the estimation of an outlier factor, which is based on the number of data outliers in the trajectory.…”

“…Our proposal mainly differs from the works by [ 20 , 22 ] since they use RNN and LSTM networks, while we propose fully connected and convolutional models additional to relative spatial relationships that can recognize outlier trajectories. Also, [ 21 , 23 ] models are based on autoregressive and neighborhood-based models that do not necessarily capture non-linear relationships between sequence measurements. The complex model presented in [ 25 ] considers maps with multiple trajectories, while we consider single trajectories.…”