Implementation of Real-Time Lane Detection on Autonomous Mobile Robot

摘要

This paper describes the implementation of a learning-based lane detection algorithm on an autonomous mobile robot. More specifically, it explores the implementation of the Ultra Fast Lane Detection algorithm for real-time application on the SEATER (Single-passenger Electric Autonomous Transporter) prototype using a camera, with an emphasis on optimizing its performance on a Jetson Nano platform. Multi-core processing on the CPU and GPU has been implemented in this paper to optimize the application program and avoid delays or buffering. Preliminary experiments have been conducted to evaluate the algorithm's performance in terms of data processing speed and accuracy using two datasets: a public dataset for the outdoor environment and another dataset collected from the indoor area of the Workshop Building at BRIN Bandung. The experimental results show that after conversion to TensorRT, the algorithm achieves significantly improved performance on the Jetson Nano platform compared to the ONNX model, with processing speeds of approximately 101 ms using CULane and 105 ms using TuSimple-about 22 times faster than the previous model. While the algorithm demonstrates good accuracy on the public outdoor dataset, its performance falls short on the indoor dataset. Future work should focus on transfer learning and fine-tuning to improve indoor lane detection accuracy.