Source-linked AI summary
DAWN: Vehicle Detection in Adverse Weather Nature Dataset
Mourad A. Kenk, Mahmoud Hassaballah
TL;DR
Adverse weather can substantially impair camera-based vehicle detection, while existing evaluations rely heavily on synthetic or limited real-world imagery. The paper introduces DAWN, a benchmark of real-traffic images spanning four adverse-weather categories with bounding-box annotations. DAWN provides a basis for examining vehicle detection across varied natural weather and traffic conditions.
Problem
Adverse weather degrades camera visibility and vehicle detection, while existing datasets provide limited coverage of challenging real-world weather conditions.
Method
The paper constructs DAWN from 1K real-traffic images across fog, rain, snow, and sand conditions, with diverse scenes and object bounding-box annotations.
Results
DAWN covers varied vehicle categories, sizes, orientations, illumination, positions, occlusions, traffic flows, and adverse-weather conditions for evaluating detection and classification.
Takeaways & Limitations
DAWN enables researchers to examine vehicle-detection aspects in natural adverse weather relevant to autonomous vehicles and intelligent transportation safety applications.
Abstract
from arXiv · showhide
Recently, self-driving vehicles have been introduced with several automated features including lane-keep assistance, queuing assistance in traffic-jam, parking assistance and crash avoidance. These self-driving vehicles and intelligent visual traffic surveillance systems mainly depend on cameras and sensors fusion systems. Adverse weather conditions such as heavy fog, rain, snow, and sandstorms are considered dangerous restrictions of the functionality of cameras impacting seriously the performance of adopted computer vision algorithms for scene understanding (i.e., vehicle detection, tracking, and recognition in traffic scenes). For example, reflection coming from rain flow and ice over roads could cause massive detection errors which will affect the performance of intelligent visual traffic systems. Additionally, scene understanding and vehicle detection algorithms are mostly evaluated using datasets contain certain types of synthetic images plus a few real-world images. Thus, it is uncertain how these algorithms would perform on unclear images acquired in the wild and how the progress of these algorithms is standardized in the field. To this end, we present a new dataset (benchmark) consisting of real-world images collected under various adverse weather conditions called DAWN. This dataset emphasizes a diverse traffic environment (urban, highway and freeway) as well as a rich variety of traffic flow. The DAWN dataset comprises a collection of 1000 images from real-traffic environments, which are divided into four sets of weather conditions: fog, snow, rain and sandstorms. The dataset is annotated with object bounding boxes for autonomous driving and video surveillance scenarios. This data helps interpreting effects caused by the adverse weather conditions on the performance of vehicle detection systems.
I. INTRODUCTION
Vehicle detection for autonomous driving must combine real-time speed with high accuracy, but adverse weather degrades camera visibility and remains insufficiently addressed by existing datasets and detectors. DAWN responds with a real-world benchmark covering multiple adverse conditions and traffic environments.
- Motivation: Adverse weather reduces camera visibility and can impair accurate vehicle detection, tracking, and traffic-scene understanding.The paper identifies fog, rain, snow, dust, sleet, and low light as challenging conditions.
- Requirements: Real-time detection and high detection accuracy are both required for autonomous-vehicle and traffic-monitoring applications under adverse weather.The paper connects real-time response to vehicle controllers and accuracy to reliable traffic-object detection.
- Existing methods: Existing detector studies often report fast detection without improving accuracy, while high-resolution systems may fail to exceed 30 frames per second under difficult lighting.The cited methods illustrate a speed–accuracy and deployment challenge for adverse-weather applications.
- Dataset gap: Available vehicle datasets do not generically combine challenging conditions such as winter weather, sleeting rain, and dust storms.The paper contrasts this gap with datasets collected from traffic, road, web, and drone cameras.
- DAWN contribution: DAWN introduces real-world adverse-weather images across fog, rain, snow, and sandstorms, with diverse traffic environments and annotated vehicle categories.The annotations target intelligent surveillance, traffic monitoring, and self-driving applications.
II. BACKGROUND
The paper evaluates vehicle-detection methods across adverse-weather conditions and multiple traffic-object categories, including vehicles, pedestrians, and cyclists.
- Evaluation scope: Performance analysis covers vehicle categories and the person category across different adverse-weather conditions.The listed vehicle classes are car, bus, truck, motorcycle, and bicycle.
A. Summary of the available vehicles image datasets in literature.
The reviewed datasets summarize existing vehicle-detection resources, while the paper emphasizes that many established traffic datasets are limited to ordinary-weather imagery and conventional scene variation.
- Dataset overview: The literature review organizes vehicle datasets by collection source, including traffic-surveillance, on-road, web, and drone cameras.Table I summarizes the available datasets and uses L, O, and C for lighting variation, occlusion, and crowdedness.
- KITTI dataset: KITTI provides 7,481 training images, 7,518 testing images, six classes, and 17,607 ground-truth bounding boxes in normal daylight traffic scenes.Its scenes span freeways, rural zones, and urban environments with lighting variability.
1) KITTI dataset
MS-COCO offers more challenging scene understanding than KITTI through large-scale, complex scenes annotated across 80 classes, but its traffic settings remain ordinary-weather scenarios.
- MS-COCO dataset: MS-COCO contains large-scale complex scenes annotated for 80 classes and is widely used by state-of-the-art deep-learning models.Its traffic objects and environmental scenes address general settings for normal weather.
2) The MS-COCO dataset
Existing traffic datasets often focus on normal-weather scenes, while DAWN highlights the greater weather and traffic-environment variability missing from state-of-the-art comparisons.
- 2) The MS-COCO dataset: DAWN contrasts with state-of-the-art traffic datasets through more extreme weather conditions and varied traffic environments.The comparison includes Rain & Snow and BDD datasets.
3) DAWN dataset:
DAWN addresses the limited availability of real-world adverse-weather data by providing diverse natural traffic images across four weather categories and varied driving environments. Its annotations support evaluating vehicle detection and classification under cross-generalization conditions.
- 3) DAWN dataset:: DAWN was introduced because existing adverse-weather datasets rely on synthetic weather or only a few real-world images.The paper identifies uncertainty about algorithm performance on uncontrolled images acquired in the wild.
- 3) DAWN dataset:: DAWN covers natural traffic scenes across fog, snow, rain, and sand conditions for cross-generalization evaluation.The dataset includes variation in vehicle category, size, orientation, pose, illumination, position, and occlusion.
- 3) DAWN dataset:: The dataset spans congested and normally moving traffic across motorways, highways, urban roads, and intersections in multiple regions.These settings are intended to represent weather and traffic-environment variation.
- 3) DAWN dataset:: Images were gathered through Google and Bing searches using adverse-weather and traffic-related queries, then filtered by humans in the loop.Selected images were required to comply with the applicable Google, Bing, and Flickr terms of use.
III. CONCLUSION
The paper concludes by presenting DAWN as a dataset for vehicle detection in adverse weather. Its distinctive coverage is intended to support investigation of vehicle-detection issues relevant to autonomous vehicles and intelligent transportation safety.
- III. CONCLUSION: DAWN is a novel dataset for vehicle detection in heavy fog, rain, snow, and sandstorms.The authors present its distinctive characteristics as an opportunity to examine previously unexamined vehicle-detection aspects relevant to autonomous vehicles and ITS safety.