Computer Vision for Healthy Driving Detection Using Convolution Neural Network

Abstract

Driving involves a rigorous act where the driver controls the operation of a motor vehicle. There have been few deployments of healthy driving applications, while some of these applications are machine learning applications some are program-driven applications. Nigeria as a developing country has little or no trained datasets for healthy driving, therefore this research will be charged with collecting local data for driving events to be trained. The datasets were collected as images. These images were extracted for driving events braking, safe driving, and speeding. The images were locally collected for Nigeria driving settings and computer vision techniques were applied to the data. The machine learning algorithm used to evaluate the model is Convolution Neural Network, the editors used for image labelling and coding the system are Jupyter notebook and VS Code. Python programming language and its libraries were also used. The classification results for model loss, accuracy, validate loss and validate accuracy and the performance of the model is 0.99 or 99%, based on this the last epoch was recorded and the loss was 0.03 or 3%.This classification result proved that the data collected from Nigeria is trainable. The trained data can be used by researchers all over the world working on safe and healthy systems in Nigeria for driving. The result also presented a convolution neural network as an algorithm suitable for healthy driving detection using computer vision. The predicted values for the three driving events were all positive. The three driving events were all detected perfectly while running the parallel testing without being perverse.