| Abstract |
In the contemporary context, the imperative to strengthen security and safety measures has become increasingly evident. Given the rapid pace of technological advancement, the development of intelligent and efficient surveillance solutions has garnered significant interest, particularly within the realm of smart city (SC). Surveillance systems have been transformed with the emergence of edge technology (ET), the Internet of Things (IoT), and deep learning (DL) to become key components of SC, notably the domain of face recognition (FR). This work introduces a smart surveillance car robot based on the ESP32-CAM micro-controller, coupled with a FR model that combines DL models and traditional algorithms. The Haar-Cascade (HC) algorithm is employed for face detection, while feature extraction relies on a proposed convolutional neural network (CNN) and predifined DL models, VGG and ResNet. While the classification is made by two distinct algorithms: Naive Bayes (NB) and K-nearest neighbors (KNN). Validation experiments demonstrate the superiority of a composite model comprising HC, VGG, and KNN, achieving accuracy rates of 92.00%, 94.00%, and 96.00% on the LFW, AR, and ORL databases, respectively. Additionally, the surveillance car robot exhibits real-time responsiveness, including email alert notifications, and boasts an exceptional recognition accuracy rate of 99.00% on a custom database. This ET surveillance solution offers advantages of energy efficiency, portability, remote accessibility, and economic affordability. |