چكيده به لاتين
Detection of anomalies in moving object trajectories is an important issue in the field of monitoring. An anomaly can be an event that does not match an expected pattern; the anomaly detection problem is divided into two general field of the anomaly detection of a trajectory and the anomaly detection between a set of trajectories.
Another issue that can be raised in the field of anomalies is the problem of prediction of anomalies. The purpose of prediction of anomaly is to detect anormalies before it can occur, which can be done to achieve various goals such as traffic control and route suggestion.
Considering that the detection of anomalies between the trajectories is an important issue and is usually applicable in important fields such as monitoring and control, nevertheless, little work has been done in this field, and most of the algorithms proposed for this field have weaknesses and They suffer from limitations, including that most of these methods are not designed for online applications; in most of these methods, there are many parameters that can be difficult to balance.
In this research, different methods of detection of anormalies were first investigated and then a high level comparisons and overall classification were presented on these methods.
Then and because the raw data is not suitable for the processing, a preprocess is performed on the data. Then, in the next step, a nearest-neighbor method is proposed to quantify the anomaly of reach road segment which has several advantages, including the independence of the distribution of data and the ability to generate output both in the form of a label and in the form of a score. In the next step, an online method for detecting and predicting traffic anomalies has been proposed. Highlights of this section introduce two types of inherent anomalies and induction anomalies that consider hidden but influential factors in traffic anomalies to predict anomalies. At the end of this research, various experiments have been carried out using a standard data set on the proposed algorithms, and the proposed algorithms are compared with other methods in this field and the results of these experiments are mentioned.
