| In order to further standardize the management of toll roads, and coordinate with the work of controlling transport of out-of-gauge goods as well, part of toll roads in 29 provinces across the country implement weight-calculating charge. For the purpose of reducing vehicle weight, truck drivers take some illegal operations, such as jumping, taking the s-shaped line and pressing the roadside, while trucks are passing the weighing platform. It has become an increasingly concerned question that how to find hidden toll evasion and provide condemnatory evidence. Therefore, it has important practical value to study some kind of method possible to detect the beating of vehicles when they are passing the weighing platform.Based on basic theories of video detection, this paper compared some common detection methods, such as background subtraction, frame difference and optical flow method through experiments. According to the results, background subtraction was utilized to detect existence of vehicles in video images, and frame difference was used to detect movement of vehicles. Then, license plate precise location algorithm was researched. License plate image was initially located through color feature-based method, and further accurately located by texture entropy and standard deviation-based method. After the license plate was correctly located, the track detection algorithm based on license plate centroid difference was put forward. Considering that the daily traffic of trucks was heavy sometimes, the track detection algorithm above was improved to save hard disk space by filtering static vehicle images and only processing and storing.the dynamic ones. The experimental results showed that the improved algorithm can accurately extract vehicle trajectories and substantially save storage space.Whether the vehicle was beating can be judged through analyzing the amplitude and slope characteristic of the trajectory that was fit by using track point average method. This trajectory's peak position was found out by backward difference, and its peak value and the maximum positive slope were calculated. By comparing calculated values of peak value and the maximum positive slope with the thresholds given, whether the vehicle was beating can be judged. The experimental results showed that it can detect jumping of vehicle with high precision by extracting and analyzing the characteristic of trajectory. The method presented by this paper is feasible. |
