A Steering Strategy for Self-Driving Automobile Systems Based on Lane-Line Detection

Steering strategy is an essential task for self-driving automobile systems. However, studies on this problem have not yet achieved satisfactory results and typically cause navigation tasks to be difficult. Therefore, this paper proposes a novel steering strategy based on the lane-line detection model. First and foremost, the row-based selecting strategy and CNN-based extraction were adopted to anticipate lane-line markers from the images captured from a front-view monocular camera. Next, the lane-line detection model output is utilized to estimate the next destination for the self-driving automobile, and then we converted the model to TensorRT pattern with Float16 format. Depending on the result of the lane-line detection model, we designed a strategy to control the steering wheel through a DC Servo motor. Last but not least, the whole algorithm is deployed on the golf cart to perform navigation tasks. The experimental result proves that our model achieves approximately 50 frames per second (50 fps) on our laptop with GTX 1650 graphic card during the testing stage and can work with satisfactory performance on the HCMUTE campus.