Pedestrian Detection using HOG + LBP + SVM Algorithm

In this paper, we implemented a pedestrian detection system using the combined HOG + LBP + SVM algorithm. This approach integrates Histogram of Oriented Gradients (HOG) features, Local Binary Patterns (LBP) features, and a Support Vector Machine (SVM) classifier. The HOG feature extraction algorithm calculates gradient magnitude and orientation histograms across image cells, effectively capturing pedestrian shape and contour characteristics. For implementation, we typically use sliding windows with blocks of 16x16 pixels and 8x8 pixel cells, creating feature vectors that represent human silhouette patterns. The LBP feature extraction component analyzes texture information by comparing each pixel with its neighbors, generating binary patterns that encode local texture variations. In code implementation, we often apply uniform LBP patterns with 8 sampling points and radius 1, creating texture descriptors that complement the shape information from HOG features. The SVM classifier, trained on labeled pedestrian datasets, learns to differentiate between pedestrian and non-pedestrian patterns. We typically employ a linear kernel SVM with parameters optimized through cross-validation. During detection, the system extracts combined HOG-LBP features from image regions and feeds them to the SVM for classification. This multi-feature combination enhances detection performance by leveraging both shape (HOG) and texture (LBP) information, resulting in improved accuracy and robustness in identifying pedestrian targets within images. The implementation typically involves OpenCV libraries for feature extraction and scikit-learn or libsvm for classifier training and inference.