Research Of Exception Object Detection Problem In Trajectory Data

In recent years,with the rising popularity of GPS,sensor networks and wireless communication applications,more and more the behavior of the trajectory data is collected and stored in the application server,such as campus identification data,data bus card,large enterprise employee card data,and so on.These valuable data contains the rich knowledge,through the behavior of the objects recorded data,which can be a lot of analysis and research,including its trajectory classification,clustering,exception object detection,etc.The exception object detection is an important direction of the trajectory data research.Trajectory data expressed an object in a certain time and place for an activity.Similarity to the action of a same group object trajectory,found that was different from most of the objects on the behavior of objects is of great significance.Trajectory data behavior can be divided into three categories,which are based on time,based on the location and based on the event.Anomaly detection is to detect outliers,it is to find out its behavior is very different in the process of the expected objects.Normally,anomaly detection are used in trajectory data,in trajectory data,independent objects are composed of several interconnected points.While the existing exception object detection in trajectory data problems are mostly based on the trajectory of continuous time series data of anomaly detection,commonly used method is to extract sequence characteristics,which formation sequence pattern,through the pattern matching to detect abnormal objects.For anomaly detection based on event trigger trajectory data,although there has been a certain research achievements at present,but in the actual application scenario,because of the uncertainty of the event,not a completely suitable for anomaly detection algorithm of various kinds of application scenarios.In this paper,we study the behavior of the trajectory data,which is based on the behavior of the event and there is a group relation between objects,and which based on the above two points,so this paper introduce the object trajectory similarity computing the exception object detection algorithm.Trajectory similarity calculation is very important of exception object detection algorithm in this paper.To accurate and efficient calculating behavior path similarity,this paper puts forward four kinds of behavior trajectory and the calculation method of similarity algorithm,which based on sliding window,respectively,based on integral sliding window algorithm,based on the single weight of the hash table algorithm and the algorithm based on double hash table.In the four kinds of algorithms,based on integral sliding window algorithm is according to sliding window algorithm on the basis of the modified,by putting the object as a whole and improve the efficiency of algorithm.Based on a single hash table algorithm and the algorithm based on double hash table are based on the strategy of hash table design,the two algorithms on the overall performance is better than the previous two kinds of algorithms.This paper adopts a university campus one real experiment data,the comprehensive evaluation the performance of the four algorithms.Experimental results show that the trajectory similarity in behavior,if the data volume is small,so the efficiency of the four algorithms were similar,but with a gradual increase in the amount of data,two algorithms based on the strategy of hash table efficiency than the two algorithm based on sliding window strategy with high efficiency.At the same time,this paper also analyzed the influence of trajectory similarity threshold parameter time influence on four algorithm,can be seen from the results,appropriate time threshold on the efficiency of algorithm is improved.This paper also analyzes the top-k exception object,to verify the effectiveness of the proposed in this paper introduce the trajectory similarity calculation the correctness of the exception object detection.