Vision-based gender and age classification

Abstract

Gender and age classification are tasks in which humans excel. It will be more convenient and helpful for surveillance and human-AI interaction if we can do this task automatically with a computer. Many image processing solutions have been proposed to classify users' appearances on different platforms. Most studies are focus on the face's image. However, those methods may not be feasible in actual use as humans may not be cooperative and the images' resolution may not be clear enough for analysis. This paper will introduce a convolution neural network (CNN) to train the network for recognizing the gait shape of different gender and age groups with a python program. Unlike the face's image processing, the gait-based method can analyse the characteristic of humans at a longer distance without any cooperation from the object. This efficient solution can be widely used in medical, criminal, and economical points of view. The CNN is trained with accuracy about 80% or higher.