Development of Microcontroller based Computer Vision Using CMSIS-NN to Detect Driver Drowsiness

Low, Wei Han (2021) Development of Microcontroller based Computer Vision Using CMSIS-NN to Detect Driver Drowsiness. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text Low Wei Han.pdf Restricted to Registered users only Download (2MB)

Abstract

The power and processing capabilities of microcontroller has progress so much that now it has the ability to perform machine learning on it. As the life in a city become more modern, more vehicle is travelling in the road every day and the accident that caused by the drowsiness of a driver has become common. In this proposal, a driver drowsiness detection system using microcontroller-based computer vision is proposed. Real time application of driver eye monitoring with high accuracy and speed can be achieved by using Convolutional Neural Networks (CNN). The driver drowsiness detection system will detect the drowsiness of the driver by monitoring the eye activity of the driver. The algorithm will process the input image and detect whether the driver drowsy or not. This only hardware required for this project is the STM32F746G-Discovery board and STM32F4DIS-CAM camera module. The software used for the development of this program for this project is Mbed Cli and Virtual Box which is used to run the Ubuntu linux. In this thesis, a base program is built before the proposed system is developed to test the feasibility of the system. The model that deployed in the project are trained using the Tensor Flow Lite model maker provided by TensorFlow. The system only require a very little amount of power for it to run normally, so it is very energy efficient compared to other system that are developed to run on high power device such as computer. Lastly, the developed prototype system are able to differentiate the images of a opened eye and closed eye with 70% accuracy.