Characterization, Detection, and Segmentation of Work-Zone Scenes From Naturalistic Driving Data

This paper elucidates the automatic detection and analysis of work zones (construction zones) in naturalistic roadway images. An underlying motivation is to identify locations that may pose a challenge to advanced driver assistance systems (ADAS) or autonomous vehicle navigation systems. Researchers first present an in-depth characterization of work-zone scenes from a custom data set collected from more than a million miles of naturalistic driving data. They then describe two machine learning algorithms based on the ResNet and U-Net architectures. The first approach works in an image classification framework that classifies an image as a work-zone scene or non-work-zone scene. The second algorithm was developed to identify individual components representing evidence of a work zone (signs, barriers, machines, etc.). These systems achieved an 𝐹0.5 score of 0.951 for the classification task and an 𝐹1 score of 0.611 for the segmentation task. They further demonstrate the viability of their proposed models through saliency map analysis and ablation studies. To their knowledge, this is the first study to consider the detection of work zones in large-scale naturalistic data. The systems demonstrate potential for real-time detection of construction zones using forward-looking cameras mounted on automobiles. Such a system can be incorporated with ADAS to assist drivers in navigating through challenging environments such as construction zones, making those areas safer for commuters.