The main objective of autonomous driving is to make vehicle understand about its surrounding environment. CARLA simulator provides great simulated platform for testing autonomous vehicle with different AI techniques. Initially, short-est path between user defined source and destination is found among all the possible paths using A star algorithm. Further, instance segmentation is used to conduct road detection to deter-mine drivable area. For this scenario, Detectron2 is implemented with achieved accuracy of 98.3%. To locate obstacles in its way, object detection and classification is performed using YOLOv4. The proposed YOLOv4 model has achieved mAP accuracy of 92.6%. Finally, tracking of moving cars is accomplished which employs DeepSORT algorithm. Therefore, proposed research work has a unique capability for conducting important tasks of autonomous vehicle such as finding shortest path, road detection, object detection, classification and tracking in single module.
Object Detection and Tracking for Autonomous Vehicle using AI in CARLA
2022-09-23
3017594 byte
Conference paper
Electronic Resource
English
A Waypoint Tracking Controller for Autonomous Vehicles Using CARLA Simulator
Springer Verlag | 2022
|No Cost Autonomous Vehicle Advancements in CARLA through ROS
British Library Conference Proceedings | 2021
|