Research On Surveillance Video Analysis-based Visibility Detection And Early Warning System For China Highway

In recent years, due to the environmental deterioration, more and more low-visibility weather, caused by haze, dust, highlight and so on, has been emerging increasingly, posing serious threats to the safety of highway transportation. As a result, it’s urgent to fully monitor the roadside visibility conditions, and to comprehensively warn and dispose under low visibility circumstances, for China highway system. Comparing to the traditional meteorological visibility-detecting instruments, which are often blamed as expensive, limited sample space, difficult for installation and maintenance, the surveillance video frames recorded from the densely-deployed video camera devices along highway road also contain plenty of roadside visibility information, and propose a novel concept for highway visibility detection. In this dissertation, a surveillance video-based method for highway visibility detection by utilizing video frames collected from roadside cameras was approached, and atop of which, an information system of roadside visibility detection, alarm and disposal was developed. The major works of this dissertation are as follows:(1) System framework design. By considering the requirements of visibility monitoring, alarm and disposal from transportation industry, a framework of surveillance video analysis-based visibility detecting and warning system for China highway was sketched. This framework is composed mainly by multiple lower end subsystems and a solely upper end subsystem. The former subsystems are used for surveillance video analysis and visibility calculation above roadside video camera facilities, while the latter one is in charge of visibility monitoring, low-visibility warning and disposition, and visibility information publication to the end users.(2) Roadside visibility detection method. The major steps of this method is as follows, firstly a baffle of fixed specification and color patterns are installed in sight of the densely distributed surveillance cameras, then the candidate visibility values are calculated according to reference image distortion analysis on these images based on the classical visibility theory and the video image analysis technology. Experimental results show that, this method exhibits a high detection precision, and meets the accuracy requirements of roadside visibility detection for China highway system.(3) Development of the visibility detection and warning system. Within this system, the lower end subsystems are deployed on embedded industrial control computers, and have a plugin-based software structure; while the upper end subsystem, developed on JAVA EE technology, has a layered structure with layers for data entities, business logics, and front Web pages. Demonstrational applications show that, this system can continuously and fully detect the visibility along the highway with 15% detection error, and can quickly publish low visibility events, and therefore has a significant application value in transportation industry.