Abstract:The scarcity of comprehensive datasets in the traffic light detection and recognition domain and the poor performance of state-of-the-art models under hostile weather conditions present significant challenges. To address these issues, this paper proposes a novel approach by merging two widely used datasets, LISA and S2TLD. The merged dataset is further processed to tackle class imbalance, a common problem in this domain. This merged dataset becomes our source domain. Synthetic rain and fog are added to the dataset to create our target domain. We employ Fourier Domain Adaptation (FDA) to create a final dataset with a minimized domain gap between the two datasets, helping the model trained on this final dataset adapt to rainy and foggy weather conditions. Additionally, we explore Semi-Supervised Learning (SSL) techniques to leverage the available data more effectively. Experimental results demonstrate that models trained on FDA-augmented images outperform those trained without FDA across confidence-dependent and independent metrics, like mAP50, mAP50-95, Precision, and Recall. The best-performing model, YOLOv8, achieved a Precision increase of 5.1860%, Recall increase of 14.8009%, mAP50 increase of 9.5074%, and mAP50-95 increase of 19.5035%. On average, percentage increases of 7.6892% in Precision, 19.9069% in Recall, 15.8506% in mAP50, and 23.8099% in mAP50-95 were observed across all models, highlighting the effectiveness of FDA in mitigating the impact of adverse weather conditions on model performance. These improvements pave the way for real-world applications where reliable performance in challenging environmental conditions is critical.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is the poor performance of traffic light detection and recognition under adverse weather conditions. Specifically, the researchers face the following challenges: 1. **Insufficient Datasets**: The existing field of traffic light detection and recognition lacks comprehensive datasets, which makes it difficult for models to generalize to various practical scenarios. 2. **Impact of Adverse Weather**: The existing state - of - the - art (SOTA) models show a significant performance drop under adverse weather conditions such as rainy or foggy days, because these conditions reduce image quality and thus affect the accuracy of object detection. 3. **Class Imbalance**: There are fewer samples of yellow traffic lights, resulting in poor recognition ability of the model for this type of traffic lights. To solve these problems, the authors propose a new method. They combine two widely - used datasets, LISA and S2TLD, and introduce synthetic rainy and foggy images to create the target - domain dataset. Then, they use **Fourier Domain Adaptation (FDA)** technology to narrow the gap between the source domain and the target domain, enabling the model to better adapt to adverse weather conditions. In addition, they also explore **Semi - Supervised Learning (SSL)** technology to make more effective use of the available data. ### Specific Solutions: - **Dataset Merging and Pre - processing**: Merge the LISA and S2TLD datasets and solve the class imbalance problem through multiple enhancement techniques (such as horizontal flipping, random brightness - contrast adjustment, affine transformation, and blurring). - **Generate Synthetic Adverse - weather Images**: Use existing methods to add synthetic rain and fog effects to create the target - domain dataset. - **Fourier Domain Adaptation (FDA)**: Replace the low - frequency components of the source image and the target image to generate FDA - enhanced images, thereby reducing the inter - domain gap. - **Semi - Supervised Learning (SSL)**: Use the YOLOv6 model to train 50% of the labeled data and generate pseudo - labels for the remaining 50% of the data to further improve the model performance. ### Experimental Results: The experimental results show that the FDA - enhanced images are superior to the non - FDA - enhanced images in multiple evaluation metrics (such as mAP50, mAP50 - 95, Precision, and Recall). For example, the YOLOv8 model has a 5.1860% increase in Precision, a 14.8009% increase in Recall, a 9.5074% increase in mAP50, and a 19.5035% increase in mAP50 - 95. ### Summary: Through these improvements, this research significantly improves the performance of traffic light detection under adverse weather conditions, providing more reliable technical support for autonomous driving systems and other related applications.

TLDR: Traffic Light Detection using Fourier Domain Adaptation in Hostile WeatheR

A Domain Adaptation Framework for Short-term Traffic Prediction.

Domain Adaptation based Object Detection for Autonomous Driving in Foggy and Rainy Weather

Snowy Scenes,Clear Detections: A Robust Model for Traffic Light Detection in Adverse Weather Conditions

Lidar Light Scattering Augmentation (LISA): Physics-based Simulation of Adverse Weather Conditions for 3D Object Detection

Domain Adaptive Object Detection for Autonomous Driving under Foggy Weather

Prior-Based Domain Adaptive Object Detection for Hazy and Rainy Conditions

Multiple Adverse Weather Conditions Adaptation for Object Detection via Causal Intervention

Multi-Target Domain Adaptation via Unsupervised Domain Classification for Weather Invariant Object Detection

Object Detection in Adverse Weather for Autonomous Vehicles Based on Sensor Fusion and Incremental Learning

Object Detection in Adverse Weather for Autonomous Driving through Data Merging and YOLOv8

Progressive Domain Adaptive Object Detection Based on Self‐Attention in Foggy Weather

Comprehensive Augmented Domain Adaptation for Image Segmentation under Rainy Conditions

Enhancing Autonomous Vehicle Perception in Adverse Weather through Image Augmentation during Semantic Segmentation Training

DAGL-Faster: Domain adaptive faster r-cnn for vehicle object detection in rainy and foggy weather conditions

DA-RAW: Domain Adaptive Object Detection for Real-World Adverse Weather Conditions

Image-Adaptive YOLO for Object Detection in Adverse Weather Conditions

Vehicle Detection and Tracking in Adverse Weather Using a Deep Learning Framework

WARLearn: Weather-Adaptive Representation Learning

Deep Learning with Attention Mechanisms for Road Weather Detection

FogAdapt: Self-Supervised Domain Adaptation for Semantic Segmentation of Foggy Images