Abstract:
With the rise of artificial intelligence in the 21st century, a host of industries were set to be revolutionized and the automotive industry is very much a part of it. A few decades ago, autonomous cars were not a possibility due to insufficient processing power, memory, and storage space. However, all of that has changed and now we have cars which either possess complete self-driving capabilities or a high-level of driver-assistance features. Self-driving cars are not only the future, but they also promise to reduce, if not eradicate, accidents caused due to factors such as human negligence or traffic-law violations. The concept also promises a future where driving a car would not remain a human-dependent responsibility and it would rather develop into a single-click action. The project is focused on developing a prototype of a scaled-down version of an autonomous car within a controlled environment. It would be capable of driving in road lanes using lane detection, observing, and obeying traffic-lights using classification, detect and avoid obstacles, and drive without any human intervention. This will be achieved by using deep-learning algorithms in addition to computer vision techniques. The primary hardware modules include an RC car and a custom-designed road-track. In addition to this, an array of sensors and a single-board computer (Raspberry Pi) will be used to sense the surrounds and perform computation using algorithms respectively. The neural networks will be trained on a dataset collected using the custom-designed road-track and the accuracy of the model will then be tested on the subject road-track.
