Pedestrian Recognition based on Human Semantics and PCA-HOG

Abstract

In real monitoring scenarios, pedestrian semantics, such as gender and clothing type, is very important for pedestrian retrieval and pedestrian recognition. Traditional pedestrian semantics attribute recognition algorithm adopts manual feature extraction and cannot express the association between pedestrian semantics features. This paper proposes a pedestrian semantics recognition method based on improved AlexNet convolution neural network to obtain pedestrian semantics features. Vector. At the same time, a large number of experiments show that HOG descriptors have a good effect in pedestrian recognition, but the number is too large. In this paper, PCA-HOG descriptors are used to express pedestrians and obtain low-dimensional PCA-HOG eigenvectors. Finally, PCA-HOG feature vectors and pedestrian semantic feature vectors are joined together, and LR model is used to predict pedestrian recognition. Compared with traditional methods, the algorithm is simpler, more practical and has higher recognition accuracy.