Object and Traffic Light Recognition Model Development Using Multi-GPU Architecture for Autonomous Bus
The autonomous vehicle is both an exciting yet complex field to dig in these past few years. Many have ventured out to develop Level 4 Autonomous Vehicle but up to this point, many issues were still arising about its safety, perception and sensing capabilities, tracking, and localization. This paper aims to address the struggles of developing an acceptable model for object detection in real-time. Object detection is one of the challenging areas of autonomous vehicles due to the limitations of the camera, lidar, radar, and other sensors, especially during night-time. There were various datasets and models available, but the number of samples, the labels, the occlusions, and other factors may affect the performance of the dataset. To address the mentioned problem, this study has undergone a rigorous process of scene selection and imitation to deal with the imbalance dataset, applied the state-of-the-art YOLO architecture for the model development. After the development process, the model was deployed in a multi-GPU architecture that lessens the computational load on a single GPU structure and was tested on a 12-meter fully electric autonomous bus. This study will lead to the development of a usable and safe autonomous bus that will lead the future of public transportation.