HybridEI: Smartly Face Detection System in Resource Constrained Edge Environment

With the digital revolution of the Internet of Things (IoT), data is rapidly emerging at the network’s edge, where it is collected using artificial intelligence (AI) and processed in a variety of applications, including cybersecurity, image recognition, and human tracking. Despite this, facial expression recognition (FER) remains a challenging problem in computer vision. Inconsistencies in the sophistication of FER and variations among expression classes are often ignored in most facial expression detection systems due to the various expressions of emotion and unpredictable environmental influences. An HybridEI (Hybrid Edge Intelligence), a smartly face detection system trained in a more powerful cloud environment and inferenced at the resource-constrained edge environment, is proposed for FER in this study. Face classification complexity, memory minimization, and computational time were all well addressed by our proposed HybridEI approach. With an accuracy of 82.78% and 1.27 frames per second (FPS), the experimental results on FER-2013 dataset demonstrated the effectiveness and superiority over other state-of-the-art (SOTA) methods. Furthermore, both the memory capacity and computational time were significantly decreased.