Desarrollo de un sistema domótico con reconocimiento facial y chatbot para gestión y control de seguridad de una residencia.

Abstract

In today's modern world, home security has become one of the most important issues to solve properly, so the primary initiative behind IoT is to facilitate the connection of various household appliances to a local network and provide a unique identity to enable the exchange of data and get to establish functions to provide security to the residence. The construction of software and hardware that integrate security systems have had a great exhaustive evolution, biometric systems is a sample of this, so much that have been used for a long time for its effectiveness, as is the case of the classic authentication with keys, then with the arrival of the fingerprint, card access, constant monitoring with video surveillance, sensor networks. Today, artificial intelligence. AI has been incorporated more naturally into the daily life of the population and the operations of organizations. Artificial intelligence is evolving to bring countless benefits to all productive sectors such as healthcare, retail, manufacturing, law enforcement and education. Recently, advances in surveillance systems have occurred thanks to the different devices of the, IoT internet of things. This work presents a system that allows the control and management of security in a residence through facial recognition, using deep learning. Deep Learning, in an embedded device called. Jetson Nano, which incorporates a USB camera for the creation of images, facial recognition for surveillance, as well as an electric plate for access to the residence. The architecture for the construction of this project was based on free software using the Linux operating system in the Ubuntu distribution, the high-level programming language Python that allowed the construction of the system, implementing libraries for computer vision such as OpenCV and for facial detection and recognition Dlib with the face_recongnition api that incorporates pre-trained models with convolutional neural networks for detection and classification algorithm "knn" improved with an accuracy of 99.38%. The Jetson.Gpio digital signal handler library was used for access. A generic natural language-based chatbot was used to control surveillance and manage access to prevent anyone from entering the residence. After developing the system and assembling the devices, the necessary tests were carried out to measure the efficiency of the system in different lighting conditions, these defined according to the variations of time in the day, which determine the direction that takes the light, the test individual was located at different distances, these defined according to the logic of the people who have at the time of being located in front of the camera, both facial recognition for surveillance and access. The results show that the detection and recognition algorithm works effectively in a low-cost integrated system.