A New Smart Home Intruder Detection System Based on Deep Learning
Hiba Hameed Ali, Jolan Rokan Naif, Waleed Rasheed Humood
Abstract
The security of home doors has become one of the necessities in this era. The Internet of Things (IoT) technology has also entered into building the smart home. Therefore, it has become necessary to develop a facial recognition system that can be implemented on IoT devices. This study presented a method to recognize faces using the efficientnet-b4. Transfer learning with fine-tuning was used here due to the small dataset size and high accuracy (accuracy of Top-1= 82.9% and accuracy of Top-5 = 96.4%) of EfficientNet-B4 and it has fewer parameters (19.5 M) than the previously known model and this is what we are looking for in order to implement it on the Raspberry Pi. After training and saving the model, it is converted into a lightweight model and transferred to the Raspberry to distinguish faces. The results showed that the model had an accuracy of 97%, despite the fact that the collected images were taken in different lighting, different places, and different facial expressions.