Age Group Recognizer based on Human Face Supporting Smart Digital Advertising Platforms

Smart digital advertising platforms have been widely employed in public areas in big cities. An age group recognizer is indispensable to support these platforms in providing relevant advertisements for each audience. These platforms also demand a recognizer that can run properly at the minimum on a CPU device to degrade the budget during system procurement. This study proposes an age group recognizer based on human faces to predict the age group of the audience’s face using an efficient architecture containing a light backbone. This work offers a residual mini multi-level module integrating two grouped convolution layers with diverse frequency levels to extract exclusive facial features maintained by residual operation. In order to improve the feature map’s quality, a deep lite attention module is proposed, consisting of the deep channel and lite spatial attention part. The architecture generates few parameters with cheap operation and achieves competitive performance on the benchmark datasets. In addition, the architecture integrated with face detection, as a recognizer, can perform fast on a CPU configuration in real-time with 144 frames per second.