Part-wise pedestrian gender recognition via deep convolutional neural networks

摘要

Gender prediction is a challenging field covering automated pedestrian attributes analysis. Machine learning approaches are in use to closely predict correct gender. In order to improve the accuracy of prediction, a deep convolutional neural network is proposed to analyze the pedestrian gender. In the methodology, the pedestrian's images are parsed with the help of existing deep de-compositional neural network. The parsed images with removed background are then divided into full body and upper body images. Later, these two types of images are given as an input to the proposed fine-tuned convolutional neural network model. The full body input images are classified into frontal-view, back-view, and mixed view gender categories. Using upper body clothing, images are sub-classified into eight categories. The proposed approach proved to have better prediction performance with respect to different sub-classifications.