New spectral eye video database SPEED revolutionises eye-tracking
Techniques to acquire spectral data have been static for a long time - until now. Exciting and novel spectral video technologies are emerging, allowing us to extract increasingly dynamic knowledge from light. Using a spectral video device in eye-tracking, computational spectral imaging and eye-tracking researchers from the University of Eastern Finland have created a novel - first of its kind - combined spectral video/spectral image database: the SPectral Eye vidEo Database, SPEED.
The potential of spectral imaging technologies has been shown in many fields, such as medicine, life science and industry where the spectrum has been used, for example, to differentiate healthy and unhealthy tissue or separate real and counterfeit objects. However, these technologies are often too slow to capture spectral images from dynamic objects. This created an evolutionary need for the extension of spectral imaging to the recording of spectral video. In a spectral video, each frame of the video stream contains a spectral image cube. In this way, spectral video systems capture spatial, spectral and temporal information, and combine the advantages of spectral imaging and video capture.
A recent PhD thesis completed at the University of Eastern Finland explored the potential of such spectral video technology by testing an existing spectral video system and challenging it in practical eye-tracking application.. As a result, the study developed protocols for the characterization of spectral video that may serve as future guidelines on spectral video systems evaluation in this emerging field. In addition, the study used a spectral video camera to capture spectral videos of a fast-moving object: the human eye. As a very concrete outcome of the study, the SPectral Eye vidEo Database, SPEED, was born.
The unique database, SPEED, was motivated by the challenges faced in eye-tracking, especially under harsh conditions. When users wear eyewear or make-up, reflections and extreme eye positions interfere with eye-tracking technologies. The team proposed and demonstrated that spectral signatures can be exploited to create new approaches for imaging, training, analysis and interpretation of eye-tracking data, especially in harsh conditions. SPEED consists of over 180 spectral videos, it continues to grow and it is publicly available to all researchers (e.g., for teaching-related needs or creating and testing new methods of dynamic eye analysis via spectra). Although published only in 2016, the database has already attracted attention from the scientific community and industry. Aside from its use in eye tracking, SPEED also has the potential to be utilised in other eye-related research areas, such as medicine, biometrics, and eye and vision research.