Gender and Age Estimation at Distance in Smart Cities Surveillance: A cascaded Deep Learning-Based Approach

Smart cities are becoming a reality with a big number of mounted cameras on each street. Those cameras provide a huge amount of data that humans cannot process it. The main goal of those cameras is to detect criminals, detect sick people, and find lost children. In this paper, we propose an intelligent gender and age estimation system for smart city surveillance. The recent improvement of convolutional neural networks for computer vision tasks makes it suitable for the proposed intelligent system. However, the distance separating the camera from the target person can be a hard challenge that must be solved. The idea of this work is to detect faces in real scene images and estimates gender and age by analyzing facial landmarks. To do that, we propose to combine a deep convolutional generative adversarial network, the viola jones algorithm, and a convolutional neural network to generate predictions on gender and age. The generative adversarial network was used to enhance the resolution of the images, the viola jones algorithm was used to detect and crop faces in the images, and the convolutional neural network was used to generate predictions on the gender and the age of the detected face. The proposed approach was evaluated on the IMDB-WIKI dataset. The obtained results prove the efficiency of the proposed approach.