Deployment of Real-Time Object Recognition in Raspberry Pi with Neural Compute Stick for Blind and Deaf People

This study details the process of building a system to identify objects that make use of Raspberry Pi, neural networks, and several sensors. To make a flexible and portable setup, it is necessary to combine a Raspberry Pi board with components such as the Pico W, ESP32Cam, L293D Driver, vibration motor, ultrasonic sensor, as well as Bluetooth module. Two object detection models such as YOLO and SSD are used in this research study and it is required to modify them using TensorFlow Lite so that it can run efficiently on Raspberry Pi. This work discusses about the ways to gather and label data for training the proposed model and also goes over the possibility of using pre-trained models for quicker deployment. Because of its modular design, the proposed system can easily accommodate the addition of new sensors or actuators.