Advanced Driver Assistance Systems (ADAS) generally utilize cameras to provide limited automation functions to enhance driver safety.ADAS uses computer vision (CV) to extract vehicle surrounds, lane boundaries, drivable regions, and nearby objects. ADAS systems fail, however, when the vehicle is operating in adverse conditions (e.g., obscured lane lines). We introduce a new ADAS CV method that was evaluated in snowy weather and lane line occlusion scenarios by recognizing tire tracks in the snow. This approach was previously developed using classical machine learning techniques (ML) but has now been expanded to include a variety of convolutional neural network (CNN) models. Using an instrumented automation research vehicle, a custom dataset was collected. A data preparation pipeline was constructed for data labeling and model training. The CNN models outperform the classical ML model in detecting tire tracks on key metrics such as Intersection over union (IoU), precision, and recall, at the expense of real-time compute speeds in frames per second (FPS). Essentially, we have demonstrated that this method works as an end-to-end pipeline for detecting tire tracks since it doesn’t necessitate any feature engineering and is a feasible way to expand the operational design domain of current ADAS products
Development of Computer Vision Models for Drivable Region Detection in Snow Occluded Lane Lines
Machine Learning and Optimization Techniques for Automotive Cyber-Physical Systems ; Kapitel : 21 ; 591-623
2023-03-27
33 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
British Library Conference Proceedings | 2022
|British Library Conference Proceedings | 2022
|SAE Technical Papers | 2022
|British Library Conference Proceedings | 2022
|