Real-Time Emotion and Gender Recognition Using Mini-Xception CNN

In this work, we describe a cutting-edge approach to face recognition in near real-time. Our system can detect human faces, identify the person who is displaying a face, categorise them as a male or female through the use of image data, assigns emotion type based on facial expressions, etc. Face detection is accomplished with OpenCV-based technology and an ensemble of highly trained convolutional neural networks (CNN's) performed with standard dataset training data. The end result is a system capable of more accurately detecting human faces in near real-time video (including images) and web services than conventional systems when detecting the emotion expressed on the person's facial feature. The back end of our system has been developed using the popular Flask framework, thereby allowing our software to run as an independent application without regard to the operating environment(s) it will run on, as well as providing easy accessibility for third parties to integrate it into their existing and future systems. As such, we have developed a technology which can be used to build human-robot interface systems, allowing humans to communicate via facial expressions with their robot assistants and provide the most accurate means of identifying and categorising the emotions being displayed on human faces.