Traffic signs are critical for maintaining the safety and efficiency of our roads. Therefore, we need to carefully assess the capabilities and limitations of automated traffic sign detection systems. Existing traffic sign datasets are limited in terms of type and severity of challenging conditions. Metadata corresponding to these conditions are unavailable and it is not possible to investigate the effect of a single factor because of simultaneous changes in numerous conditions. To overcome the shortcomings in existing datasets, we introduced the CURE-TSD-Real dataset, which is based on simulated challenging conditions that correspond to adversaries that can occur in real-world environments and systems.

Dataset:  Google form for dataset access

Paper: https://arxiv.org/abs/1908.11262

We generated synthesized video sequences with a professional game development tool Unreal Engine 4 (UE4). We utilized UE4 to synthesize video sequences because of three main advantages. First, objects in UE4 have metadata including position, size, and bounding box, which eliminates the manual labeling process. Second, environmental conditions including weather, time, and lighting can be fully controlled in UE4. We can change specific environmental conditions and levels while fixing other parameters to perform controlled experiments. Third, UE4 is relatively user friendly and is supported by a large developer community. We started data generation process by creating a virtual city that included roads, street lamps, traffic signs, and background environment. We added global lighting sources to make virtual environment more realistic. We generated a car object and linked that object with two functions, one for obtaining the location and the other one for showing the object on the screen. We attached a camera object to the car object, designed a driving path, created a path follower, configured the car speed, and placed the vehicle on the designed path. To match the number of real- world sequences, we generated 49 simulated sequences, which leads to a total of 98 reference videos denoted as challenge-free sequences.

Dataset:  Google form for dataset access

Paper: https://arxiv.org/abs/1902.06857

In this dataset, we investigate the robustness of traffic sign recognition algorithms under challenging conditions. Existing datasets are limited in terms of their size and challenging condition coverage, which motivated us to generate the Challenging Unreal and Real Environments for Traffic Sign Recognition (CURE-TSR) dataset. It includes more than two million traffic sign images that are based on real world and simulator data. We benchmark the performance of existing solutions in real-world scenarios and analyze the performance variation with respect to challenging conditions. We show that challenging conditions can decrease the performance of baseline methods significantly, especially if these challenging conditions result in loss or misplacement of spatial information. We also investigate the effect of data augmentation and show that utilization of simulator data along with real-world data enhance the average recognition performance in real-world scenarios. 

Dataset:  https://ghassanalregib.com/cure-tsr/

Paper: https://arxiv.org/abs/1712.02463

We utilized our patented lab-on-a-headset platform to perform automated light rexlex test and generated our RAPD dataset. Lab-on-a-headset is an ultra portable device with on- board processing capability, which enabled us to capture and preview patient videos during the clinical study. We predefined the test sequence by connecting to the headset and utilizing the graphical user interface that provides full control over the test stimuli. Tested subjects were stimulated with the automated light sequences and their pupillary reactions were recorded simultaneously as high definition streams. In Fig. 3, we show sample images captured with the introduced headset. We used the infrared frames to assess relative afferent pupillary defect in this study. We obtained approvals from the Institutional Review Board committees of Emory University and Georgia Institute of Technology. RAPD conditions of the subject were determined by clinicians with manual swinging flashlight test as well as neutral density filter test. The RAPD dataset used in study included 22 subjects, half of which correspond to a control group without RAPD whereas the other half have positive RAPD. Four out of ten males and seven out of twelve females have positive RAPD. Average age of the participants is around 52 years for no RAPD subjects and 56 years for RAPD positive subjects with a standard deviation of approximately 14 and 10, respectively.

Dataset:  Not publicy available

Paper:https://arxiv.org/abs/1908.02300, https://arxiv.org/abs/1905.08886