Deep Neural Network Based Roadwork Detection for Autonomous Driving

Sebastian Wullrich; Nicolai Steinke; Daniel Goehring

cs.ROApr 2, 2026

Deep Neural Network Based Roadwork Detection for Autonomous Driving

Authors: Sebastian Wullrich, Nicolai Steinke, Daniel Goehring

Organizations: Dahlem Center for Machine Learning and Robotics, Freie Universität Berlin

Abstract

Road construction sites create major challenges for both autonomous vehicles and human drivers due to their highly dynamic and heterogeneous nature. This paper presents a real-time system that detects and localizes roadworks by combining a YOLO neural network with LiDAR data. The system identifies individual roadwork objects while driving, merges them into coherent construction sites and records their outlines in world coordinates. The model training was based on an adapted US dataset and a new dataset collected from test drives with a prototype vehicle in Berlin, Germany. Evaluations on real-world road construction sites showed a localization accuracy below 0.5 m. The system can support traffic authorities with up-to-date roadwork data and could enable autonomous vehicles to navigate construction sites more safely in the future.

Figures & tables

Panel of Figure 1 — Fig. 1 : Roadwork objects indicating road construction sites in Germany [ 9 ] : (a) Barrier; (b) Traffic cone; (c) Left-facing vertical panel (“pass left”); (d) Right-facing vertical panel (“pass right”)

Fig. 2 : Dataflow of the roadwork detection system

Fig. 3 : Red area: illustrates the current measurement of roadwork dimensions utilizing detected roadwork objects as the vehicle passes (moving from bottom to top, passing on the right); Gray area: undetected portion representing the full extent of the actual roadwork if all its objects were detected

Panel of Figure 4 — Fig. 4 : Differing shapes of final roadwork outputs: (a) Unaltered version recorded during driving; (b) Refined version using only convex hull outer points

Panel of Figure 5 — Fig. 5 : Average confusion matrices from 6-fold cross-validation of the final YOLO model: (a) A confidence threshold of 0.001; (b) A class-specific threshold of 0.75 for barriers and 0.7 for other classes

Fig. 6 : False positive example detected by the final model

Fig. 7 : False negative examples detected by the final model

Metric	Barrier	Cone	Vertical Panel pass left	Vertical Panel pass right
Accuracy	$0.885\pm 0.015$	$1.000\pm 0.000$	$0.916\pm 0.010$	$0.974\pm 0.004$
Precision	$0.990\pm 0.005$	$1.000\pm 0.000$	$0.992\pm 0.003$	$0.989\pm 0.005$
Recall	$0.763\pm 0.030$	$0.950\pm 0.112$	$0.745\pm 0.031$	$0.880\pm 0.016$
F1-score	$0.861\pm 0.020$	$0.971\pm 0.066$	$0.850\pm 0.020$	$0.931\pm 0.010$

TABLE I : Mean and standard deviation per class of the different metrics across all six folds of the cross-validation of the final YOLO model with the new confidence thresholds

Panel of Figure 9 — Fig. 8 : Examples where the LUX sensor contours differ from the Velodyne point cloud

Fig. 10 : Comparison of roadwork measured by the system (red) and Geoportal Berlin [ 4 ] entry (green), registered half a block away from the actual position, therefore failing to show the intersection blockage

Fig. 11 : Vertical panel incorrectly associated with other panels across the road, causing a single roadwork to be formed due to the system’s inability to account for road curvature

Explore similar work

Framework and Multi-modal Dataset for Roadwork Zone Detection and Geo-localization
Jul 5, 2026Zhiran Yan, Yutong Xin, S Shyam Shenoi +2Autonomous Driving Datasets Autonomous Driving Perception
Institute of Innovative Mobility (IIMo), Technical University Ingolstadt of Applied Sciences, Ingolstadt, 85049, Germany. · Technical University of Munich, Garching, 85748, Germany. · Fraunhofer Institute for Transportation and Infrastructure Systems IVI, Ingolstadt, 85051, Germany.
WorkDrive: Roadwork Chain of Causation for Autonomous Driving
Jul 16, 2026Tianyi Jiang, Wen Zhang, Sihan Yang +2VLMs for Autonomous Driving Autonomous Driving Perception
Peking University · Beijing, China · Xiaomi EV
Autoware in Construction: Gap Analysis and LiDAR Perception Toward Off-Road Autonomous Driving
Oct 3, 2026Yu Otsuki, Teja Emmey, Sena Matsushita +2Autonomous Driving Autonomous Driving Perception
Department of Civil and Environmental Engineering, University of California, Irvine (UCI), Irvine, CA, USA · TIER IV, Inc., Shinagawa, Tokyo, Japan

Jul 5, 2026J/K move · Enter open · S save

Framework and Multi-modal Dataset for Roadwork Zone Detection and Geo-localization

Zhiran Yan, Yutong Xin, S Shyam Shenoi, Rui Song+1

Institute of Innovative Mobility (IIMo), Technical University Ingolstadt of Applied Sciences, Ingolstadt, 85049, Germany. · Technical University of Munich, Garching, 85748, Germany. · Fraunhofer Institute for Transportation and Infrastructure Systems IVI, Ingolstadt, 85051, Germany.

Autonomous Driving Datasets Autonomous Driving Perception

Open paper AskPDF arXiv

Autonomous vehicles often rely on high-definition (HD) maps for navigation; however, these maps are not frequently updated and often lack semi-static information, such as temporary roadwork zones, which can significantly alter the road network. This limitation underscores the urgent need for an accurate global position of roadwork zones. However, the absence of publicly available datasets for evaluating roadwork zone detection and geo-localization models has hindered the development of reliable autonomous driving systems. To address this challenge, we propose the Roadwork Zone Detection and Geo-localization (RZDG) dataset, which includes both simulated and real-world data, providing multimodal sensor inputs along with comprehensive annotations. The dataset supports multiple perception tasks, including image semantic segmentation, 3D object detection, and object geo-localization. In addition, we introduce a tracker-based roadwork zone detection and geo-localization (RZDG) pipeline, an extension of AB3DMOT, for accurate object geo-localization in roadwork zones. We benchmark our approach on the RZDG dataset, demonstrating its effectiveness in detecting roadwork zones and transforming object positions from the local coordinate system to the global coordinate system. A prediction is considered a true positive (TP) if its estimated position falls within one meter of the ground truth. Our experimental results show that our approach achieves high accuracy on both real and simulated data. Specifically, we report: Precision: 0.565 (real) / 0.615 (simulated) Recall: 0.898 (real) / 0.809 (simulated) F1-score: 0.597 (real) / 0.665 (simulated).

Abstract

Explore similar work

Framework and Multi-modal Dataset for Roadwork Zone Detection and Geo-localization

WorkDrive: Roadwork Chain of Causation for Autonomous Driving

Autoware in Construction: Gap Analysis and LiDAR Perception Toward Off-Road Autonomous Driving