| Bad weather and traffic abnormality usually bring great threatens to the safety of highway. The management department of highway needs to grasp the real-time weather and traffic information of the areas in charge, so as to make timely and correct decisions. However, the weather and traffic information are obtained by the analysis of data provided by the sensors and other hardware devices at present. The installation and maintenance cost is high, and the service time and environmental condition may affect its accuracy.This thesis makes full use of the highway surveillance architecture. The bad weather (including fog, rain and snow) detection algorithm and traffic abnormalities (including congestion, parking and retrograde) detection algorithm based on highway surveillance video are studied. The weather and traffic information are obtained by the intelligent analysis of video data come from the camera. This algorithm combines many advanced technologies of the image processing, computer vision and pattern recognition field. The visual features of fog, rain and snow appear in the image are extracted and analyzed in order to achieve detection and alarming of the bad weather. Meanwhile, the moving vehicles are tracked and their behaviors are analyzed in order to achieve the detection and alarming of the traffic abnormalities.This thesis analyzes the visual features of the fog, rain and snow show in the surveillance videos. Aimed at the blurring effect of the fog, we propose a Canny edge based fog detection algorithm. For the rain and snow detection, considering the characteristic of the surveillance video that the resolution is small and the noise interference is big, we abandon the traditional approaches which analyze the dynamic characteristic of raindrops and snowflakes, and turn to analyze the visual feature of the road in the rainy and snowy day. Aimed at the reflective and dull property of the rainy road, we propose a slippery road evaluating method combining image reflection with sharpness, to implement the rain detection from the perspective of judging the road is slippery or not. Aimed at the color character of the covering snow, we propose a covering snow detection method based on the snow color model, to implement the snow detection from the perspective of judging the road is covering snow or not.For the traffic abnormality detection, this thesis firstly analyzes several common technologies on foreground detection and background reconstruction, and makes a comparison through experiments. This algorithm combines background subtraction with the statistical model based background reconstruction and updating method to implement vehicle detection and tracking. The approximate vehicle speed, traffic flow and direction are estimated. We make a comprehensive analysis and judgment using the estimated information to implement the detection of parking, retrograde and congestion on the highway.This thesis has a varied video data using as experimental dataset. The statistic and analysis of the experimental result illustrate the effectiveness on bad weather detection and traffic abnormality detection of the algorithm proposed in this thesis. A practical intelligent analysis system has been implemented based on this algorithm. The system is compatible with the video surveillance platform. It can analyze not only local historical videos, but also online real-time videos in the way of multi-channel parallel polling. It is an important component of the highway information service architecture. With full function and high flexibility, its application will effectively improve the capability of the highway on management and emergency response. |
PDF Full Text Request