| Abstract: |
Machine vision techniques for road scene analysis typically require the determination of a region of interest (ROI) to enable efficient and rapid image processing. The ROI is a designated area within an image that is processed for tasks such as lane detection, traffic sign recognition, and object detection. However, previous research on ROI determination has primarily focused on lane detection, excluding other critical road elements that provide essential driving information, such as warning signs, directional indicators, and speed limits. In this study, we propose an adaptive ROI determination algorithm that extends the conventional ROI beyond lane regions to include road markings information relevant to driving. The proposed method utilizes semantic segmentation based on the Cityscapes dataset, a widely used benchmark for autonomous driving applications, to identify road regions comprehensively. To evaluate the effectiveness of road recognition, we conducted experiments by varying the number of segmentation classes. Specifically, we tested three configurations: (1) utilizing all classes available in the dataset, (2) selecting 10 classes relevant to road attributes, and (3) considering only the road class. The experimental results demonstrate that the road-only classification model achieves a high recognition rate for straight, curved, and rural roads, and expressways. |