2025 ASEE Annual Conference & Exposition

This work presents advancements in computer vision methodologies aimed at enhancing the safety and adaptability of autonomous vehicles in diverse driving environments. This study addresses key challenges such as real-time processing, environmental variability, and scene understanding by refining Mask R-CNN’s object detection and segmentation capabilities and introducing a novel scene classifier. Mask R-CNN improvements enable precise identification of critical objects such as pedestrians, vehicles, and traffic signs. At the same time, the scene classifier dynamically adjusts detection parameters to optimize performance across urban, rural, and highway contexts under varying weather and lighting conditions.

The integration of these technologies improves real-time responsiveness and computational efficiency, which is crucial for dynamic autonomous driving applications. Evaluation metrics demonstrate significant gains in detection accuracy and processing speed, including mean Average Precision (mAP), Intersection over Union (IoU), and frame processing time. Preliminary results also highlight the effectiveness of data augmentation techniques and multimodal sensor data in mitigating challenges posed by adverse weather and ambiguous scenes.

This research contributes to developing robust, context-aware autonomous systems that enhance intelligent transportation networks' safety, reliability, and efficiency. Future directions include leveraging edge computing and advanced AI architectures to improve decision-making processes and achieve Level 5 autonomy.