Smart Hat for the blind with Real-Time Object Detection using Raspberry Pi and TensorFlow Lite
Matshehla Konaite, Pius Adewale Owolawi, Temitope Mapayi, Vusi Malele, Kehinde O. Odeyemi, Gbolahan Aiyetoro, J. S. Ojo
Abstract
The purpose of this paper is to introduce a system that uses the combination of working aid blind techniques to help blind people to have better navigation against objects, also to be aware of road and traffic lights signs using real-time image-processing. This study aims at improving the processing speed of object detection by introducing the latest Raspberry Pi 4 module, which is more powerful than the previous versions. As a result, the Single-Shot Multibox Detector MobileNet v2 convolutional neural network on Raspberry Pi 4 using TensorFlow Lite 2, is employed for object detection. A model called SSD MobileNet v2 320x320, which is trained on the COCO dataset from the TensorFlow model zoo, which is used in this system. Also, the traffic light signs classification is achieved via machine learning techniques. With all modules attached on the hat, video frames captured from the piCamera module are analyzed and searched for known objects by the Raspberry Pi module and playing the pre-recorded sound of the detected object(s) at an approximate distance away from the user. MobileNet uses 3 x 3 depthwise separable convolutions, which uses between 8 to 9 times lesser computing resources compared to the standard convolution at a small cost in accuracy. Thus, this makes the device suitable for the resource constraint devices such as the Raspberry Pi. The model achieves about 5 frames per second on the Raspberry P 4.