In this paper, a fully automated method for calibrating the camera and obtaining dimensions and speed of vehicles is presented. In this method, at first, vanishing points and the focal length of the camera are obtained, according to the directions of the cars in the ini
More
In this paper, a fully automated method for calibrating the camera and obtaining dimensions and speed of vehicles is presented. In this method, at first, vanishing points and the focal length of the camera are obtained, according to the directions of the cars in the initial frames. After detecting moving vehicles, their 3D bounding box are created using the vanishing points. In order to deal with the perspective, the bounding box of each vehicle is projected on a hypothetical road and then to have their real dimensions in meter, the metric coefficient (pixel-to-meter conversion) is obtained. This calculates the coefficient; a dominant car is detected and according to its metric dimensions, the pixel to meter coefficient is computed. Projecting the vehicle on the road surface and the use of the metric coefficient provides the possibility of expressing the actual speeds and dimensions of the vehicles in each frame. However, it may have some errors. To increase the accuracy of the results, these parameters are aggregated along the vehicle's path, and some histograms are made up for the speed and dimensions of each vehicle. Then the maximum of these histograms is reported as new values of speed and dimensions for each vehicle. This will improve the accuracy. Creating histograms for each vehicle requires tracking of the car in multiple frames. For this purpose, a fast algorithm is presented. Comparing the results of the proposed method with previous methods indicates higher processing speed and better response. The average error of dimension estimation is 1.4%, and the error of speed estimation is 1.1 km/h. The average processing speed for testing videos in MATLAB is about 3.5 frames per second.
Manuscript profile
