This RTSD: Russian Traffic Sign Images Dataset was created for training and testing the algorithms of traffic sign recognition. Frames are obtained from widescreen digital video recorder which captures 5 frames per second. Frame resolution is from 1280×720 to 1920×1080. Frames are captured in different seasons (spring, autumn, winter), time of day (morning, afternoon, evening) and in different weather conditions (rain, snow, bright sun).

Motivation

Recently, the development of control systems for self-driving cars has been progressing rapidly. A critical component of these systems is the algorithm for recognizing road signs. Beyond autonomous vehicles, sign recognition is also used in driver assistance systems and to automate road maintenance services. Modern methods of object recognition in images rely on deep learning models. To create a high-quality detection model, a well-labeled dataset is essential. However, the process of annotating datasets is both expensive and time-consuming, requiring meticulous manual work and verification to account for human errors.

Synthetic data offers a solution by simplifying data collection. It can be generated quickly, at no cost, without annotation errors, and in virtually unlimited quantities, significantly reducing the cost of obtaining data. The task of recognizing road signs inherently involves highly unbalanced classes. Many classes of signs are underrepresented in training samples, complicating the model training process. Previous research has demonstrated that using realistic synthetic datasets with augmentations of rare classes of road images can enhance traffic sign recognition models.

Dataset description

The authors explore the recognition of both common and rare road signs, emphasizing that the importance of rare signs is equal to that of common ones. Firstly, they enhance the current markup of the Russian traffic signs dataset in a semi-automatic mode, adding 9,000 new signs: 4,000 to the test set and 5,000 to the training set. Secondly, they conduct an experimental evaluation of modern methods for the classification and detection of road signs. Instead of using a combined classification method, the authors employ a single neural network trained on a mixture of real and synthetic data. To boost the performance of road sign detection and classification, they utilize stochastic weight averaging. This approach results in a significant improvement in recognition metrics for both rare and common signs.

This dataset is designed for training and testing traffic sign recognition algorithms. The frames are captured using a widescreen digital video recorder at a rate of 5 frames per second, with resolutions ranging from 1280×720 to 1920×1080. The frames are taken in various seasons (spring, autumn, winter), at different times of the day (morning, afternoon, evening), and under diverse weather conditions (rain, snow, bright sun). This dataset exceeds other public traffic sign datasets in terms of the number of frames, sign classes, physical signs, and images of signs. The sign labeling process is conducted in two stages. In the first stage, tracks of physical objects are identified in sequential frames. In the second stage, indistinguishable signs are discarded, and each physical sign is assigned a class.

The problem of recognizing road signs is unsolved for a large number of classes of signs, since there are many rare classes of signs. For systems that recognize road signs, the ability to quickly adapt in the event of a new type of road signs is important. With the emergence of a new class of signs, it is difficult to collect a sufficient number of real training examples. In the presence of synthetically generated images of this sign, the ability to classify a new type of sign can be quickly introduced into the system. High-quality synthetic samples allow us to solve the problem with missing classes and data collection with examples of these signs. The quality of the detectors can also be improved with synthetic signs. In this task, the consistency of the appearance of the sign with the background is especially important. The correct positioning of the new synthetic road sign is also very important.

Real image	Image with additional signs

Note: Сlasses of the specified Russian road signs comply with GOST requirements.