Vehicle classification and detection based coarse data for warning traffic jam in VietNam
Van-Tuyen Dinh, Ngoc-Diep Luu, Hoang-Hon Trinh · 2016
This paper describes a new method to detect vehicles such as cars, motorbikes; it will be a good coarse data to analyze the traffic jam, specially in VietNam where motorbikes densely appear on the roads. From a new image, the method auto-hierarchically (automatically and hierarchically) learns and retrieves geometrical model, backgroud model, foreground objects. The geometrical model is used for reducing size of focus region, and reducing time processing. The background is automatically retrieved and updated by using Median filter method. From the background model and new image, the foreground objects are detected as candidates of Vehicles. To detect and classify the vehicle, Morphological features as area, aspect ratio, bounding box, orientation are used. From training set, the thresholds of morphological features are specified for each type of single, double or motorbikes, cars. This approach is built and finished by several well known algorithms such as line segment detection, dominant vanishing point calculation. The results shows that single, double or motorbikes, cars of vehicles are detected and classified in high accuracy and highly potential for real applications.