Investigating the Impact of Environmental Factors on Autonomous Vehicle Sensors
Open accessAbstract
The operation of autonomous vehicles requires the coordinated operation of a number of sensors to improve road safety. Data from the sensors are processed by vehicle control algorithms, which then make decisions. If there is a degradation in the efficiency of these sensors, the reliability of the whole system is affected. Different weather conditions affect the efficiency of the system. The research has identified the weather factors that affect the performance of the sensors based on a literature search. Following the literature analysis, a simulation test was carried out to investigate the extent to which the detection performance of a stereo video camera installed in an experimental autonomous vehicle is affected when part of the sensor is covered by a contaminant. The measurement was followed by a comparison of the vehicle and obstacle detection efficiency when the camera is completely clean and when part of the camera is covered.
AI snapshot
Written from the paper’s full text.
TL;DR
This study evaluates YOLOv8–YOLOv12 object detectors on a custom 558-image dataset augmented with synthetic water-droplet, mud, and sunflare perturbations to simulate sensor contamination. YOLOv10 generally gave the best precision and F1-score, while YOLOv12 showed better recall in some high-intensity scenarios.
Key findings
- Across all model sizes and perturbation types, YOLOv10 consistently achieved the highest precision and F1-score, especially under more challenging noise levels; YOLOv12 exhibited slightly better recall in several high-intensity scenarios, particularly for mud contamination and sunflare glare.
- At lower IoU thresholds (0.8 and 0.85), YOLOv12 and YOLOv10 achieved the highest number of best results, especially in medium and large parameter ranges; at IoU 0.9 and 0.95, YOLOv10 dominated across all model sizes.
- For small models under mud contamination at intensity 0.2, YOLOv10 reached precision 0.6768, recall 0.6390, and F1 0.6573, compared with YOLOv8 at 0.5639, 0.4815, and 0.5194; at intensity 1.0, YOLOv10 had precision 0.5795 but recall 0.0099 and F1 0.0195, while YOLOv12 had recall 0.0398 and F1 0.0714.
- Table 2 reports COCO mAP 50-95 values including YOLOv12n 40.6, YOLOv12s 48.0, YOLOv12l 53.7, YOLOv10n 38.5, YOLOv10s 46.3, and YOLOv10l 53.2.
Methods
- A custom dataset of 558 images at 1280 × 500 pixels was extracted from a short video recorded during a drive around the Széchenyi István University campus in Győr, Hungary, on public roads; average vehicle speed was approximately 20–25 km/h. Images were manually annotated following COCO class definitions for vehicles (cars, trucks, buses, motorcycles), pedestrians, and bicycles.
- Each image was artificially distorted using Albumentation v2.0.8 with three adversarial visual perturbations, each at five intensity levels: RandomFog (p ∈ {0.1, 0.2, 0.3, 0.4, 0.5}), Spatter (p ∈ {0.2, 0.4, 0.6, 0.8, 1.0}), and RandomSunFlare (flare_roi = (0, i, 1, i + 0.2), i ∈ {0.0, 0.1, 0.2, 0.4, 0.6}).
- YOLOv8 through YOLOv12 models in the Ultralytics framework (v8.3.221), pre-trained on COCO, were evaluated using precision, recall, and F1-score at four IoU thresholds: 0.8, 0.85, 0.9, and 0.95, yielding 45 evaluation metrics.
Limitations
- Stated: synthetic perturbations enable controlled and reproducible testing, but real-world contamination may exhibit more complex patterns; future work should include validation on real sensor data to confirm transferability.
- Evident: the dataset is small (558 images) and limited to one campus/low-speed urban driving (20–25 km/h), so results may not generalize to rural or highway conditions or higher speeds; the paper notes future plans for higher-speed testing and lane, traffic sign, and traffic light recognition.
- Evident: the abstract describes a stereo video camera contamination test, but the methods and results evaluate YOLO models on augmented images, not a physical stereo camera or real contamination; no real sensor data were used.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
1 of 2 cited sources checked · 0 supported · 0 flagged · accuracy not yet measured
- Premisecitation
Rain and snow have a significant impact on detection accuracy because water droplets and snowflakes can distort light under certain conditions, confusing the sensors and causing them to transmit less and less accurate data to the data processing system.
“Rain and snow have a significant impact on detection accuracy because water droplets and snowflakes can distort light under certain conditions, confusing the sensors and causing them to transmit less and less accurate data to the data processing system.”
Quote in the paperSubstantial claim1 of 1 sources checked- cites [8] Linnhoff, C., Hofrichter, K., Elster, L., Rosenberger, P., and Winner, H. (2022). Measuring the Influence of Environmental Conditions on Automotive Lidar Sensors. Sensors, 22. source
Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.
Experimental — accuracy not yet measured
- Findingown data
YOLOv10 demonstrated the most stable and dominant performance at higher thresholds.
“while YOLOv10 demonstrated the most stable and dominant performance at higher thresholds”
Quote in the paperSubstantial claim - Findingown data
At the lower IoU thresholds (0.8 and 0.85), YOLOv12 and YOLOv10 models achieved the highest number of best results, particularly within the medium and large parameter ranges.
“At the lower IoU thresholds (0.8 and 0.85), YOLOv12 and YOLOv10 models achieved the highest number of best results, particularly within the medium and large parameter ranges”
Quote in the paperSubstantial claim - Findingown data
As the IoU threshold increased to 0.9 and 0.95, YOLOv10 consistently dominated across all model sizes, achieving the highest number of best results independently of the number of parameters.
“However, as the IoU threshold increased to 0.9 and 0.95, YOLOv10 consistently dominated across all model sizes, achieving the highest number of best results independently of the number of parameters.”
Quote in the paperSubstantial claim - Findingown data
YOLOv10 offers the most balanced performance in terms of precision and F1-score.
“Overall, the results confirm that YOLOv10 offers the most balanced performance in terms of precision and F1-score”
Quote in the paperSubstantial claim
- Findingown data
YOLOv12 exhibited slightly better recall values in several high-intensity scenarios, particularly for mud contamination and sunflare glare.
“Notably, YOLOv12 exhibited slightly better recall values in several high-intensity scenarios, particularly for mud contamination and sunflare glare”
Quote in the paperSubstantial claim - Findingown data
YOLOv10 consistently achieved the highest precision and F1-score values across all model sizes and perturbation types, especially under more challenging noise levels.
“The analysis reveals that YOLOv10 consistently achieved the highest precision and F1-score values across all model sizes and perturbation types, especially under more challenging noise levels.”
Quote in the paperSubstantial claim - Findingown data
These trends were consistently observed across the small, medium, and large model groups.
“These trends were consistently observed across the small (Tables 345), medium (Tables 678), and large model groups (Tables 91011), emphasizing the robustness advantages of the newer model generations under adverse conditions.”
Quote in the paperSubstantial claim - Findingown data
The latest versions (YOLOv11 and YOLOv12) showed better performance trends at lower IoU thresholds.
“Furthermore, the results confirm that the latest versions (YOLOv11 and YOLOv12) showed better performance trends at lower IoU thresholds”
Quote in the paperSubstantial claim - Interpretationown data
YOLOv11 and YOLOv12's better performance trends at lower IoU thresholds suggest improved detection capabilities under looser matching conditions.
“suggesting improved detection capabilities under looser matching conditions”
Quote in the paperSubstantial claim - Interpretationown data
YOLOv10 exhibits superior robustness and localization precision under stricter detection conditions, especially when high IoU is required.
“This indicates that YOLOv10 exhibits superior robustness and localization precision under stricter detection conditions, especially when high IoU is required.”
Quote in the paperSubstantial claim - Interpretationown data
YOLOv12 and YOLOv10's ability to handle more challenging detection scenarios with higher flexibility in bounding box accuracy is reflected by their highest number of best results at lower IoU thresholds.
“reflecting their ability to handle more challenging detection scenarios with higher flexibility in bounding box accuracy”
Quote in the paperSubstantial claim - Interpretationown data
YOLOv12 shows potential benefits in situations where maintaining high recall under extreme visual degradation is critical.
“while YOLOv12 shows potential benefits in situations where maintaining high recall under extreme visual degradation is critical”
Quote in the paperSubstantial claim - Interpretationown data
YOLOv12's improved recall suggests improved sensitivity in detecting partially occluded objects.
“suggesting its improved sensitivity in detecting partially occluded objects”
Quote in the paperSubstantial claim
- Classified as probably not a substantial claim. 5
- May combine two claims. 2
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationThe tests showed that sensor performance can be reduced by up to 40-50% under the strongest distortions.
“The tests showed that sensor performance can be reduced by up to 40-50% under the strongest distortions [14].”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [14] Michaelis, C., Mitzkus, B., Geirhos, R., Rusak, E., Bringmann, O., Ecker, S.A., Bethge, M., and Brendel, W. (2019). Benchmarking Robustness in Object Detection: Autonomous Driving When Winter is Coming. arXiv. source
Probably not a substantial claim: may not pass S4 (based).
Interpretationnone givenReal-world contamination may exhibit more complex patterns than synthetic perturbations.
“a limitation is that real-world contamination may exhibit more complex patterns”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
Study design and conductnone givenFor the evaluation, a selection of YOLO models was made to ensure a representative comparison across different network sizes. Within each category, models were selected to have comparable parameter counts, ensuring a fair comparison of architectural improvements across versions.
“For the evaluation, a selection of YOLO models was made to ensure a representative comparison across different network sizes.Within each category, models were selected to have comparable parameter counts, ensuring a fair comparison of architectural improvements across versions.”
Quote in the paperNot a substantial claimMay combine two claims.
Study design and conductreasoningArtificial distortions were preferred over real contaminations to ensure full control over intensity and location of sensor degradations, enabling reproducible and safety-compliant testing scenarios.
“Artificial distortions were preferred over real contaminations to ensure full control over intensity and location of sensor degradations, enabling reproducible and safety-compliant testing scenarios.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Study design and conductnone givenAll images were manually annotated following the COCO dataset class definitions, ensuring consistency with the object categories used during model training.
“All images were manually annotated following the COCO dataset class definitions, ensuring consistency with the object categories used during model training.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S4 (based).
Study design and conductnone givenSynthetic perturbations enable controlled and reproducible testing.
“While synthetic perturbations enable controlled and reproducible testing”
Quote in the paperNot a substantial claim
The paper relies on this without support.
The artificial distortions (RandomFog, Spatter, RandomSunFlare) adequately simulate real-world sensor degradations from water droplets, dirt, and sun glare.
“To simulate typical sensor degradations encountered in real-world applications, each image was artificially distorted using the Albumentation (v2.0.8) [22] image augmentation library.”
The paper relies on this without support.
The custom dataset (558 images from campus driving) is representative of real-world traffic scenarios for evaluating detection performance.
“The footage captures various real-world traffic scenarios, including vehicles, pedestrians, and bicycles, providing a realistic environment for evaluating detection performance.”
The paper relies on this without support.
The COCO-pretrained YOLO models are suitable for detecting objects in the custom dataset without additional training.
“The YOLO object detection models implemented in the Ultralytics framework (v8.3.221),pre-trained on the COCO dataset, were tested in three scenarios, with three types of adversarial visual perturbations applied to the images.”
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
YOLOv10 does not consistently have the highest F1-score
MinorThe discussion claims YOLOv10 has the highest F1-score across all model sizes and perturbation types. In Table 7, the medium-model mud results at effect level 0.2 give YOLOv12 an F1-score of 0.6772, above YOLOv10's 0.6662.
“YOLOv10 consistently achieved the highest precision and F1-score values across all model sizes and perturbation types”
“| 0.2 | 0.5886 0.5987 0.7202 0.6335 0.7006 0.5294 0.5382 0.6196 0.5913 0.6553 0.5575 0.5668 0.6662 0.6117 0.6772 |”
Cites
Microsoft COCO: Common Objects in Context
Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P., & Zitnick, C. L. (2014) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-10602-1_48 · Cited by 45,460 · cited 1× · see citing works
“Based on the above, the COCO dataset is one of the most important databases for machine vision applications, widely used for object detection and segmentation tasks, as well as for training and evaluating neural networks [21].”
Explaining and Harnessing Adversarial Examples
Goodfellow, I. J., Shlens, J., & Szegedy, C. (2014) · arXiv · doi:10.48550/arxiv.1412.6572 · Cited by 10,381 · see citing works
Pedestrian Detection: An Evaluation of the State of the Art
Dollár, P., Wojek, C., Schiele, B., & Perona, P. (2011) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2011.155 · Cited by 3,191 · see citing works
YOLOv10: Real-Time End-to-End Object Detection
Wang, A., Chen, H., Liu, L., Chen, K., Lin, Z., Han, J., & Ding, G. (2024) · Advances in Neural Information Processing Systems 37 · doi:10.52202/079017-3429 · Cited by 3,001
Albumentations: fast and flexible image augmentations
Buslaev, A., Iglovikov, V. I., Khvedchenya, E., Parinov, A., Druzhinin, M., & Kalinin, A. A. (2020) · Information · doi:10.3390/info11020125 · Cited by 2,714 · see citing works
Microsoft COCO: Common Objects in Context
Lin, T.-Y., Maire, M., Belongie, S., Bourdev, L., Girshick, R., Hays, J., Perona, P., Ramanan, D., Zitnick, C. L., & Dollár, P. (2014) · arXiv · doi:10.48550/arxiv.1405.0312 · Cited by 2,579 · see citing works
YOLOv12: Attention-Centric Real-Time Object Detectors
Tian, Y., Ye, Q., & Doermann, D. (2025) · arXiv · doi:10.48550/arxiv.2502.12524 · Cited by 882
Perception and sensing for autonomous vehicles under adverse weather conditions: A survey
Zhang, Y., Carballo, A., Yang, H., & Takeda, K. (2023) · ISPRS Journal of Photogrammetry and Remote Sensing · doi:10.1016/j.isprsjprs.2022.12.021 · Cited by 640 · cited 1× · see citing works
“Around the turn of the millennium, developments began and, as a result, newer and newer modern driver assistance systems, or Driver Assistance Systems (DAS) technologies, were continuously introduced [4].”
YOLOv11: An Overview of the Key Architectural Enhancements
Khanam, R., & Hussain, M. (2024) · arXiv · doi:10.48550/arxiv.2410.17725 · Cited by 604 · cited 1× · see citing works
“It provided better accuracy and efficiency while maintaining real-time applicability [19].”
Benchmarking Robustness in Object Detection: Autonomous Driving when Winter is Coming
Michaelis, C., Mitzkus, B., Geirhos, R., Rusak, E., Bringmann, O., Ecker, A. S., Bethge, M., & Brendel, W. (2019) · arXiv · doi:10.48550/arxiv.1907.07484 · Cited by 324 · see citing works
Predicting the Influence of Rain on LIDAR in ADAS
Goodin, C. T., Carruth, D. W., Doude, M., & Hudson, C. R. (2019) · Electronics · doi:10.3390/electronics8010089 · Cited by 202 · cited 1× · see citing works
“The effectiveness of environmental perception is critical for the accuracy of sign recognition and lane detection, as well as for the detection of obstacles, vehicles, and people [6,7].”
Ibrahum, A. D. M., Hussain, M., & Hong, J. (2024) · Artificial Intelligence Review · doi:10.1007/s10462-024-11014-8 · Cited by 81 · cited 1× · see citing works
“This method can be used to weaken the performance of models and the efficiency of recognition, allowing weather conditions recorded in real traffic to be tested and modified [9].”
Aloufi, N., Alnori, A. S., & Basuhail, A. A. (2024) · Electronics · doi:10.3390/electronics13153063 · Cited by 45 · see citing works
Measuring the Influence of Environmental Conditions on Automotive Lidar Sensors
Linnhoff, C., Hofrichter, K., Elster, L., Rosenberger, P., & Winner, H. (2022) · Sensors · doi:10.3390/s22145266 · Cited by 44 · cited 1× · see citing works
“Radars are less sensitive in this respect, as radar waves can penetrate rain and snow [8].”
On failures of RGB cameras and their effects in autonomous driving applications
Secci, F., & Ceccarelli, A. (2020) · 2020 IEEE 31st International Symposium on Software Reliability Engineering (ISSRE) · doi:10.1109/issre5003.2020.00011 · Cited by 42 · see citing works
Yaseen, M. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2409.07813 · Cited by 38 · see citing works
Yaseen, M. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2408.15857 · Cited by 35 · see citing works
Maqsood, M., Yasmin, S., Gillani, S. A., Aadil, F., Mehmood, I., Rho, S., & Yeo, S.-S. (2022) · ISA Transactions · doi:10.1016/j.isatra.2022.11.016 · Cited by 21 · cited 1× · see citing works
“One is to make visible changes to images (image processing attacks), where weather effects are added to the system in a way that is obvious to the human eye, changing how the image looks to people and cameras [10].”
Marchisio, A., Caramia, G., Martina, M., & Shafique, M. (2022) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn55064.2022.9892612 · Cited by 16 · cited 1× · see citing works
“The other method is the use of invisible perturbations (adversarial patterns) [11].”
Remote Perception Attacks against Camera-based Object Recognition Systems and Countermeasures
Man, Y., Li, M., & Gerdes, R. (2023) · ACM Transactions on Cyber-Physical Systems · doi:10.1145/3596221 · Cited by 6 · cited 1× · see citing works
“These modifications can be digital solutions or deceptive stickers or signs placed in real space [12].”
RGB cameras failures and their effects in autonomous driving applicationsPreprint
Secci, F., & Ceccarelli, A. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2008.05938 · Cited by 2 · see citing works
Flagged first, then by citations
Cited by
Sorted by citations
A Review of Obstacle Avoidance Path Planning Technologies for Intelligent Vehicles
Min, Y. (2026) · Computers and artificial intelligence · doi:10.70267/cai.26v3n3.135154
“Therefore, through a literature review and analysis, Csikor et al. [13] identified weather factors that affect sensor performance and the extent of their impact on detection performance, thereby providing further theoretical support for the operational stability of all the components within a…”
- Authors
- Csikor, D., Hollósi, J., & Szauter, F.
- Affiliations
- Vehicle Industry Research Center, Széchenyi István University, 1. Egyetem tér, 9026 Győr, Hungary
- Széchenyi István University (HU)
- ORCID
- Csikor, Daniel
- Venue
- SMTS 2025
- Publisher
- MDPI
- Published
- 2025-10-28
- Pages
- 15
- OpenAlex
- W4415646494
- Cited by
- 0
- References
- 13
- FWCI
- 0.00
- Type
- conference paper
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Funders
- European Commission
- Topics
- Physical Sciences › Engineering › Automotive Engineering › Autonomous Vehicle Technology and Safety
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Vision and Imaging
- Physical Sciences › Engineering › Civil and Structural Engineering › Transportation Systems and Logistics
- Keywords
- autonomous vehicles, sensor degradation, obstacle detection, detection performance, sensor reliability, road safety
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-09.
Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.
Cited sources are checked by gpt-6-sol.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
