Edge4FR: A Novel Device-Edge Collaborative Framework for Facial Recognition in Smart UAV Delivery Systems

In recent years, smart UAV (unmanned aerial vehicle) delivery has become a promising solution to solve the last-mile delivery problem in smart logistics. In a smart UAV delivery system, the accurate identification of the goods receiver is a critical task. At present, using smart lockers with quick response (QR) codes is one of the most widely used solutions. However, this solution is very expensive and limited by the space available to deploy smart lockers. In contrast, using facial recognition technology for identification is a promising solution which does not need any extra equipment besides the UAV itself. However, due to the instability and the unusual shooting angle of the UAV from the air, existing facial recognition technologies often suffer the issue of low accuracy in practice. Therefore, to improve the accuracy of UAV based facial recognition, we propose Edge4FR, a Device-Edge Collaborative Framework based on face frontalization and facial recognition. Specifically, first, the facial detection algorithm based on deep learning deployed in the UAV can detect facial images frame by frame, and extract detected faces and transmit them to the nearby edge server. Afterwards, the face frontalization model trained by the generative adversarial network (GAN) deployed in the edge server can frontalize facial images. Finally, the facial recognition algorithm based on deep learning deployed in the edge server can confirm the identity by checking if the frontal facial image matches the goods receiver’s facial image registered in the delivery system. Experimental results in a real-world smart UAV delivery system demonstrate the effectiveness of the proposed framework.