4-7-2023 (CALIFORNIA) Hollywood’s portrayal of forensics may soon face criticism as researchers at the University of Maryland unveil a remarkable technique that can reconstruct three-dimensional (3D) images from the reflections captured in a person’s eyes. This groundbreaking development builds upon a neural network model known as neural radiance fields (NeRF).
The researchers proudly showcase their accomplishments on their website, displaying impressive feats such as generating a slightly blurry but identifiable 3D rendering of a Kirby toy reflected in a subject’s eye. They have also managed to create rudimentary representations of simple scenes, including a classroom.
However, it’s not yet time for TV screenwriters to claim vindication. The photographs taken for these experiments are of very high resolution, akin to those captured by a top-of-the-line smartphone camera, and they require meticulous arrangement. Additionally, specific and controlled lighting conditions are necessary for optimal results.
To illustrate the potential for improvement, the researchers attempted a more ambitious task by analyzing real-world footage from music videos featuring Miley Cyrus and Lady Gaga. Unfortunately, the results were underwhelming, yielding only obscure black shapes. They assert that one of the shapes faintly resembles an upper torso.
Despite these limitations, the work conducted by the researchers remains highly impressive and clever. According to a pre-print study detailing the project, traditional NeRF applications typically rely on multiple moving cameras to capture a scene and generate a 3D rendering. In contrast, this novel technique utilizes a single, stationary camera and focuses on the human cornea.
The researchers leverage the geometric properties of the human cornea, which are “approximately” consistent among healthy adults. This convenient characteristic enables them to estimate the number of pixels occupied by the cornea in an image and utilize this information to position the eyes accurately in space.
However, achieving precise results presents a significant challenge, especially considering the presence of colorful irises that can obscure reflections. While the researchers have devised methods to overcome these obstacles to some extent, there is still room for improvement.
While this technique has not yet reached the level depicted in crime scene investigations on popular television shows, where detailed eyeball reflections are extracted from low-quality CCTV footage, it is undoubtedly a step towards that direction.
