Traffic Signal Recognition Image Dataset

Currently, autonomous driving technology is rapidly developing, but traffic signal recognition remains a challenge. Particularly in complex road environments, traditional algorithms often make errors under varying lighting conditions and occlusions. Existing solutions demand large amounts of data and face challenges in scene coverage. This dataset aims to provide a large and diverse collection of traffic signal images to improve algorithm accuracy under various conditions. Data is collected using dash cams and fixed cameras under various weather and lighting conditions, ensuring coverage of real-world traffic scenarios. Quality control measures include multiple rounds of annotation and consistency checking, reviewed by an expert team with backgrounds in traffic engineering and computer vision. Data preprocessing follows standardized procedures, including denoising, enhancement, and cropping, ultimately stored in structured JPG format for efficient reading and processing. The dataset is of high quality, with annotation document accuracy exceeding 95%, and model robustness is enhanced through new data augmentation techniques. Compared to similar datasets, this dataset excels in data diversity and scenario coverage, surpassing 90% of products, and can effectively improve the recognition rate of autonomous driving systems' signal detection by at least 20%. Its core advantages include data richness and rarity, especially its application value in complex and low-light environments. Additionally, the dataset is highly extensible, suitable for various global traffic systems, and can support a broader range of intelligent transportation research.