Mileage Pile Detection for Vehicle-Borne Video

To realize easy positioning and convenient operation for road maintenance, proposed the mileage pile detection and classification of vehicle-borne video for pile number recognition. This method lay the foundation for subsequent mileage pile number recognition. Firstly, Contrast Limited Adaptive Histogram Equalization (CLAHE) image enhancement enhances the contrast between the mileage pile plates and background. And then, YCrCbC simplifies redundant pixel information; and Adopting the Maximally Stable Extremal Region (MSER) roughly locates mileage pile plates area. Thirdly it adapts Histograms of Oriented Gradients (HOG) extracts the pile plate features; Lastly it uses Support Vector Machine (SVM) to classify pile plates category. The experiment results indicate that classification accuracy of 100M is 86.73% and 1 km classification accuracy is 89.77%, which provides important practice for highway maintenance.