A Novel Approach in Analyzing Traffic Flow by Extreme Learning Machine Method

Abstract

The objective of this study is to detect abnormal behaviours of moving objects captured in highway traffic flow footages, classify them by using artificial learning methods, and lastly to predict the future thereof (regression). To this end, the system being the object of the design and application consists of three stages. In the first stage, to detect the moving object in the video, background/foreground segmentation method of Mixture of Gaussian (MOG), and to track the moving object, Kalman Filter-Hungarian algorithm method have been used. In the second stage, by using the coordinates of the object, such details as location, distance in terms of time, and speed of the object are obtained, and by using total pixel count data relating to the shape of the object are obtained. The software based on the specifically elaborated algorithm compares these data with the data in the table of rules set down for the road under surveillance, and generates an attribute table comprising anomalies of the objects in the video. In the last stage, however, the data included in the attribute table have been classified and predictions by the artificial learning method, Extreme Learning Machine (ELM) made.

Keywords

• anomaly classification and prediction (regression)
• artificial learning
• Extreme Learning Machine (ELM)
• traffic flow video analysis