Pedestrian Detection Program Using HOG and LBP Feature Extraction

This program is designed to detect pedestrians in images. The implementation follows a structured pipeline: First, the program extracts both HOG (Histogram of Oriented Gradients) and LBP (Local Binary Patterns) features from positive (pedestrian) and negative (non-pedestrian) training samples. The feature extraction process involves calculating gradient orientations and local texture patterns using optimized algorithms. Second, the program trains a pedestrian classifier using Support Vector Machines (SVM), where the feature vectors are normalized and fed into the SVM training module with appropriate kernel selection. Third, the trained classifier is deployed using a sliding window approach across test images, with multi-scale detection capability. Experimental results confirm that this method effectively identifies pedestrians in images and achieves satisfactory detection accuracy. This technology finds applications in various domains including video surveillance systems and intelligent transportation systems.