Lane Classification and Object Detection for Positioning of Autonomous Vehicle Based on Stereo Vision Karina Ardellia Arfian (a), Yul Yunazwin Nazaruddin (b), Vebi Nadhira (c)
a) Department of Engineering Physics
Institut Teknologi Bandung
Bandung, Indonesia
karinaardellia5[at]gmail.com
b) 1. Instrumentation and Control Research Group, Department of Engineering Physics,
2. National Center for Sustainable Transportation Technology
CRCS Building, 2nd Floor, Institut Teknologi Bandung
Bandung, Indonesia
yul[at]tf.itb.ac.id
c) Instrumentation and Control Research Group, Department of Engineering Physics
Institut Teknologi Bandung
Bandung, Indonesia
vebi[at]tf.itb.ac.id
Abstract
The development of innovation in the field of transportation is currently growing rapidly, especially in autonomous vehicle. Increasingly, research is being carried out to improve the comfort and safety factors for users. One that plays an important role in these factors is the perception system. The perception system is an overall system capable of capturing environmental data to identify locations and objects around the vehicle. Without this system, the vehicle will not be able to recognize or work according to program commands. Two important features are discussed in this paper to identify the environment, there are lane markings and surrounding objects. For the sensor, an Intel Realsense D435 stereo camera is used which is capable of receiving information in the form of position and distance to the object in front of it. The Hough Transformation method is applied for lane detection and the YOLOv3 algorithm for object detection. This paper will combine the two position data of lane markings and objects to be used as a determinant of vehicle position and orientation. It also uses a neural network algorithm that will classify the types of lane markings according to the traffic rules, so that they can decide whether the vehicle can change lanes or not.
