Histogram of Oriented Gradients (HOG) Descriptor Implementation for Grayscale Images

This documentation presents a grayscale image descriptor based on the Histogram of Oriented Gradients (HOG) method. The descriptor implementation includes gradient calculation using Sobel or Prewitt filters, orientation binning with configurable histogram parameters, and block-wise normalization for illumination invariance. The code was developed to support the research work: O. Ludwig, D. Delgado, V. Goncalves, and U. Nunes, "Trainable Classifier-Fusion Schemes: An Application To Pedestrian Detection," 12th International IEEE Conference on Intelligent Transportation Systems, St. Louis, 2009, Vol. 1, pp. 432-437. If you use this code in published research, please cite the aforementioned paper. The HOG descriptor enables more accurate pedestrian detection by capturing gradient orientation information that characterizes human silhouette contours. Key implementation features include adjustable cell sizes, orientation bin counts, and block normalization methods (L2-norm or L2-Hys). The algorithm's effectiveness has been validated in the original paper, demonstrating high accuracy rates for pedestrian detection applications. When implementing this descriptor for pedestrian detection systems, carefully review the code parameters to ensure they match your specific requirements. The implementation supports custom configuration of detection window sizes and overlapping blocks for improved feature extraction. If you encounter any issues during implementation or require technical assistance, please contact us for support.