Target Detection Algorithm Based on Multi-structure Combination HOG

In order to detect pedestrian targets of different sizes in complex environments, a pedestrian detection method based on multi-structure combination HOG (Histograms of oriented gradients) is proposed. This method generates HOG feature in series through blocks composed of cell of different structures. For new high-dimensional HOG feature, the low-dimensional HOG descriptor is extracted by PCA (Principal Component Analysis) dimensional reduction method. At last, the detection of pedestrian target is realized by using SVM (Support Vector Machine) classification tool. The experiment verified the high accuracy of multi-structure combined HOG in multi-scale pedestrian detection through the INRIADATA.