AUTONOMOUS CAR BASED ON VISION USING ADVANCED LANE DETECTION AND DEEP LEARNING TECHNIQUE IN OBJECT RECOGNITION

• Main Author: Wira Sumbaga, Aji
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/41466
• Institution: Institut Teknologi Bandung

Since the beginning invention of the car by Karl Benz (1894), it requires a driver to drive the car. Even at the start of the trial, the car driven by Karl Benz hit the wall. This indicates that it needs a good driver to drive a car so that the car can provide excellent benefits for human life. For a few centuries after the invention of the car, the focus of researchers has been specific to the passenger's comfort and convenience. Until the last few decades, the focus began to shift to advance autonomous cars. One of the main reasons for the development of autonomous cars is the existence of a bug that until now has not been resolved, the driver. Every year, 1.2 millions of people were killed by accident. With the development of this autonomous car, it is hoped that many lives can be saved every year. In addition, people with disabilities can travel without having to depend on others. The autonomous car is design based on environmental mapping using advanced lane detection and deep learning technique in object recognition. SSDLite-MobileNetV2 is the deep learning object recognition model that we use in this final project. In this final project, the autonomous car is designed on a Remote Control Car (RC Car) with a setpoint following controller. Setpoint that we use based on the middle frame of the image captured by the camera. P controller is used to controlling the car's steering using the offset value from the error between the middle of two detected line and the middle point of the camera as the input. The results of the design of an autonomous car in this thesis have produced a car that can follow the specified lane path and stop when it detects multiple objects.