The Deep Learning Method for Image Segmentation to Improve the Efficiency of Data Processing Without Compromising the Accuracy of an Autonomous Driving Country-Road Pilot System After Image Classification

Autonomous driving requires object recognition for vehicles to automatically generate a path according to their recognised environment, the conditions of which have different dims of light, from daylight to night. High-resolution images require high amounts of expensive storage as automated driving moves from urban to rural areas, where driving at night and recognising traffic signs and lights are necessary for all light conditions. Therefore, a reliable source of input, allowing for the intended performance of an autonomous driving system such as the country or rural road pilot, is necessary for adequate deployment of its functionality in its target environment. For quality criteria such as intended performance, functional reliability, safety, and correct driving behaviour are to be ensured; accuracy metrics can be a substantial contribution to the product quality criteria. Furthermore, since autonomous technology faces the challenge of being costly, thus any new innovative methods for saving costs, without comprising quality, would help to develop and enhance the chance of this developing technology being installed into more advanced automated or autonomous driving vehicles once the product safety as quality criteria can be validated on target roads. Part of this work’s limitation was that only a simulation environment was used for testing the image processing and autonomous driving accuracy models. Through research, certain algorithms were found that may be used in storage size minimisation for taking a high-resolution image; its size had to be reduced for use without compromising accuracy in the classification process. Further research in their validation may be necessary.