Image sensor
More actions
An image sensor is a semiconductor chip that converts the light focused onto it by a lens into an electronic image. It is built from a two-dimensional array of light-sensitive picture elements (pixels); each pixel turns the photons it receives into electric charge, and readout electronics convert that charge into a signal that can be stored, displayed or analyzed.[1] The two main families are the charge-coupled device (CCD) and the CMOS image sensor, also called the CMOS active pixel sensor. CMOS sensors have displaced CCDs in most visible-light imaging, from phone cameras to machine vision.[1][2]
In virtual reality (VR) and augmented reality (AR) hardware, image sensors are the core of every headset camera. Monochrome cameras on the outside of a head-mounted display supply the images for inside-out tracking and hand tracking; color cameras provide video passthrough of the real world; small infrared cameras inside the headset watch the user's eyes for eye tracking; and time-of-flight sensors measure depth. Apple's technical specifications for the Apple Vision Pro, for example, list two main cameras, six world-facing tracking cameras, four eye-tracking cameras, a TrueDepth camera and a LiDAR Scanner.[3] Engineers from the sensor maker Brillnics and Meta's Reality Labs wrote in 2023 that image sensors for consumer AR and VR devices must be optimized for computer vision algorithms, which require global shutter operation, high sensitivity, high dynamic range and very low power consumption.[4]
How it works
Photodetection and pixels
Both CCD and CMOS sensors convert light into electric charge and then process that charge into electronic signals.[1] In most sensors the light-sensitive element of each pixel is a photodiode: in their 2014 review in the IEEE Journal of the Electron Devices Society, Eric R. Fossum and Donald B. Hondongwa describe the pinned photodiode as the primary photodetector structure used in most CCD and CMOS image sensors.[5]
The two families differ in where charge is turned into a voltage. In a CCD, the charge from every pixel is shifted across the chip and passes through a very small number of output nodes, often just one, where it is converted to a voltage and sent off the chip as an analog signal. In a CMOS sensor, each pixel has its own charge-to-voltage conversion, and the chip often adds amplifiers, noise correction and digitization so that it outputs digital bits.[1] Teledyne notes that because a CCD funnels all pixels through so few amplifiers, each amplifier needs a higher bandwidth, which results in higher noise, so high-speed CMOS imagers can be designed with much lower noise.[1] In his 1997 paper "CMOS image sensors: electronic camera-on-a-chip", Fossum described CMOS active pixel sensors as competitive with CCDs in image quality while offering on-chip functionality, lower system power, lower cost and miniaturization.[6]
Color
A photodiode measures only the amount of light, not its color. Most color sensors therefore place a mosaic of color filters over the pixels. The most common layout is the one patented by Bryce E. Bayer of Eastman Kodak (US patent 3,971,065, "Color imaging array", filed 5 March 1975 and granted 20 July 1976). In its preferred form, green filters occupy every other position in both directions, about half of all pixels, to sample luminance, and red and blue filters share the remaining positions.[7] Sensors without a color filter array record a monochrome image. This is the type used in many headset tracking cameras (see below). Some designs add extra pixel types: Sony's 2012 Exmor RS sensors, for example, added W (white) pixels to the usual red, green and blue pixels to raise sensitivity in low light.[8]
Back-illuminated and stacked sensors
In a conventional front-illuminated CMOS sensor, the metal wiring and transistors sit on the side of the silicon that faces the lens, and they block part of the light on its way to the photodiode. A back-illuminated (BSI) sensor turns the structure around so that light enters through the back of the silicon substrate, with the wiring underneath. When Sony announced the development of a back-illuminated CMOS sensor on 11 June 2008 (1.75 µm pixels, five effective megapixels, 60 frames per second), it reported about +6 dB higher sensitivity and -2 dB lower random noise than its front-illuminated sensors of the same pixel size, and noted that back illumination had commonly caused problems such as noise, dark current, defective pixels and color mixing that its new photodiode structure and on-chip lens were designed to overcome.[9]
A stacked sensor goes one step further. Instead of bonding the back-illuminated pixel layer to a plain supporting substrate, it places the pixel layer on top of a second chip that carries the signal-processing circuits. On 20 August 2012 Sony announced the commercialization of its Exmor RS line (IMX135, IMX134 and ISX014), which it called the world's first stacked CMOS image sensor, with shipments starting that October.[8] Stacking also allows circuits to be placed directly beneath the pixel array, down to a separate converter for every pixel. The sensor that Reality Labs and Brillnics described in 2023 is an example: a back-illuminated pinned-photodiode layer joined to an analog-to-digital converter and 10-bit memory in each pixel by copper-to-copper hybrid bonding.[4]
Shutter and readout
Image sensors capture a frame in one of two ways. With a rolling shutter, the sensor is read out row by row, so rows are exposed at slightly different times and moving subjects, or a moving camera, can appear skewed.[10] With a global shutter, all pixels are exposed over the same interval. According to Teledyne, rolling shutters offer speed at the cost of these artifacts, while global shutters need additional electronics on the sensor, can bring higher read noise and limited frame rates, and are typically best suited to front-illuminated sensors, which have lower quantum efficiency than back-illuminated ones.[10] The Reality Labs and Brillnics authors of a 2023 sensor paper state that the computer vision algorithms used in AR and VR devices require global shutter operation.[4]
Dynamic range and HDR
Dynamic range is the span between the dimmest and the brightest light a sensor can record in one image, usually expressed in decibels. The survey of event-based vision by Guillermo Gallego and colleagues gives about 60 dB as typical for conventional frame cameras.[11] High Dynamic Range (HDR) sensors extend this in several ways. Sony's 2012 IMX135 and IMX134 offered an "HDR movie" function that sets two different exposure conditions within a single frame and combines them.[8] The Reality Labs and Brillnics digital pixel sensor uses a triple quantization scheme in which each pixel automatically chooses between a time-to-saturation measurement for bright light and two linear conversion modes for dim and medium light, reaching 127 dB.[4]
Event-based sensors
An event camera, or dynamic vision sensor, does not capture frames at all. Each pixel independently reports a change in brightness as soon as it happens, producing a stream of events that record the time, location and sign of the change. Gallego and colleagues describe their advantages as temporal resolution on the order of microseconds, a dynamic range of about 140 dB, low power use and reduced motion blur.[11] An early example is the 128x128 pixel asynchronous temporal contrast vision sensor by Patrick Lichtsteiner, Christoph Posch and Tobi Delbruck, published in 2008 with a quoted 120 dB dynamic range and 15 µs latency.[12]
History
Willard Boyle and George E. Smith conceived the charge-coupled device at Bell Labs in Murray Hill, New Jersey, on 17 October 1969. It was first intended as a memory device; it stored light-generated charge in a series of tiny capacitors and moved it across the chip. Michael Tompsett later applied it to imaging.[13] Boyle and Smith published the device in the Bell System Technical Journal in April 1970.[14] In 1975, Kodak engineer Steven Sasson built the first digital still camera using CCD technology from Fairchild Semiconductor, and in the late 1970s Sony began using CCDs in video cameras, which led to mass-produced camcorders in the 1980s.[15] Boyle and Smith shared the 2009 Nobel Prize in Physics for the invention of the CCD sensor, an imaging semiconductor circuit, and the CCD was dedicated as IEEE Milestone No. 281 on 21 October 2025.[13]
The modern CMOS active pixel sensor grew out of work at NASA's Jet Propulsion Laboratory. Eric Fossum joined JPL in 1990 and developed the CMOS active pixel sensor there; by 1993 his team saw its potential both for NASA missions and for consumer electronics.[2] His paper "Active Pixel Sensors: Are CCDs Dinosaurs?" appeared that year in the SPIE proceedings volume CCD's and Optical Sensors III (Proc. SPIE vol. 1900).[16] In 1995 Fossum, Sabrina Kemeny and two other JPL colleagues founded Photobit with an exclusive license from Caltech.[2] Photobit's PB-100 of 1999, a 352 by 288 pixel sensor, was used in the Intel Easy PC Camera and later in versions of the Logitech QuickCam.[17] Fossum told NASA's Spinoff publication that cell phones became the "killer application" for the technology, and the publication states that, outside a few niche markets, virtually all digital still and video cameras now use his invention.[2] Sony later announced back-illuminated (2008) and stacked (2012) CMOS designs.[9][8]
| Year | Milestone |
|---|---|
| 1969 | Boyle and Smith conceive the CCD at Bell Labs (17 October)[13] |
| 1970 | CCD paper published in the Bell System Technical Journal[14] |
| 1975 | Steven Sasson builds the first digital still camera at Kodak, using a Fairchild CCD[15] |
| 1976 | Bayer color filter array patent granted to Eastman Kodak[7] |
| 1990s | Fossum develops the CMOS active pixel sensor at JPL; Photobit founded in 1995[2] |
| 1997 | Fossum publishes "CMOS image sensors: electronic camera-on-a-chip"[6] |
| 1999 | Photobit PB-100 ships in PC webcams[17] |
| 2008 | Sony announces development of a back-illuminated CMOS image sensor; Lichtsteiner, Posch and Delbruck publish their event-based vision sensor[9][12] |
| 2009 | Nobel Prize in Physics awarded to Boyle and Smith for the CCD[13] |
| 2012 | Sony announces commercialization of stacked CMOS image sensors[8] |
Applications in VR and AR
Tracking cameras
Inside-out tracking systems use image sensors on the headset to map the room and follow the headset's position. Facebook's 2019 description of Oculus Insight, the tracking technology of the Oculus Quest and Oculus Rift S, says that image data from cameras in the headset helps generate a 3D map of the room, that the system combines ultra-wide-angle cameras with inertial measurement units, and that the headset cameras also detect the infrared LEDs in the controllers.[18] The Quest cameras were later used for hand tracking as well. Facebook (now Meta) described the Quest system as relying entirely on monochrome cameras, without active depth sensing or gloves,[19] and a 2020 paper by its researchers in ACM Transactions on Graphics, MEgATrack, reported real-time hand tracking from four fisheye monochrome cameras, running at 60 Hz on a PC and 30 Hz on a mobile processor.[20]
Passthrough cameras
Video passthrough shows the user a live camera view of the surroundings inside a VR headset. Because the cameras sit on the outer surface of the headset rather than at the user's eyes, the images have to be reconstructed and warped to the eye positions. The Passthrough+ system described by Gaurav Chaurasia and colleagues at Facebook in 2020 did this on the Oculus Quest using its downward-facing grayscale stereo camera pair, which runs at 30 Hz, and a Qualcomm Snapdragon 835 processor; the authors measured 49 ms between image capture and the rendering of the corresponding camera image.[21]
Later headsets added dedicated color cameras for passthrough. Meta stated in June 2023 that the Meta Quest 3 would have two 4MP RGB color cameras, a depth sensor and 10 times more pixels in passthrough than Quest 2;[22] UploadVR noted that Quest 2 passthrough had been black and white.[23] Meta's Passthrough Camera API, available from Horizon OS v74 on Quest 3 and Meta Quest 3S, gives developers access to the two forward-facing RGB cameras at up to 1280x1280 pixels (originally 1280x960) and 60 Hz, with an image capture latency of 20-40 ms.[24] When it announced the Apple Vision Pro in June 2023, Apple said the Apple R1 chip processes input from 12 cameras, five sensors and six microphones and streams new images to the displays within 12 milliseconds.[25]
Eye-tracking cameras
Headset eye tracking systems point small cameras at the eyes. Apple describes the Vision Pro system as using high-speed cameras and a ring of LEDs that project invisible light patterns onto the user's eyes.[25] Sensor makers build parts specifically for this job. On 24 August 2022 OmniVision announced the OG0TB, a three-layer stacked back-illuminated global shutter sensor with 400x400 resolution, 2.2 µm pixels and a 1.64 mm x 1.64 mm package, drawing less than 7.2 mW at 30 frames per second, with its best quantum efficiency at 940 nm in the near-infrared; OmniVision marketed it for eye and face tracking in AR, VR and mixed reality devices.[26]
Event-based sensors have also been applied to eye tracking. At IEEE VR 2021, Anastasios Angelopoulos and colleagues presented a hybrid frame and event near-eye gaze tracker with update rates beyond 10,000 Hz, which they compared with the roughly 300 Hz that conventional gaze-tracking cameras realistically reach; they reported accuracy of 0.45 to 1.75 degrees across fields of view from 45 to 98 degrees.[27] On 16 October 2023 Prophesee launched the GenX320, a 320x320 event sensor with 6.3 µm stacked back-illuminated pixels and power consumption as low as 36 µW, aimed partly at eye tracking for foveated rendering in AR and VR headsets. In the announcement, Zinn Labs said the sensor let its gaze-tracking system run below 20 mW.[28]
Depth sensors
Some image sensors measure distance instead of brightness. A time-of-flight camera uses a sensor whose pixels time or demodulate light from the camera's own emitter (see Depth sensing). Microsoft's Azure Kinect DK documentation describes its depth camera as an amplitude-modulated continuous-wave time-of-flight design with a 1-megapixel imaging chip, 3.5 µm pixels and a global shutter that improves performance in sunlight.[29] The underlying sensor was presented by Cyrus Bamji and colleagues at ISSCC 2018 as a "1Mpixel 65nm BSI 320MHz Demodulated TOF Image Sensor with 3.5μm Global Shutter Pixels and Analog Binning". Microsoft's mixed reality documentation describes the paper as covering the depth camera to be utilized in Project Kinect for Azure and the next version of HoloLens.[30]
Light field capture
A light field camera, or plenoptic camera, places a microlens array between the main lens and the image sensor. Ren Ng and colleagues at Stanford built a hand-held plenoptic camera on this principle in 2005.[31] See Microlens Arrays for how such arrays are used in AR and VR optics.
Examples
| Device or sensor | Image sensor use | Details |
|---|---|---|
| Oculus Quest | Tracking, hand tracking, passthrough | Grayscale stereo pair at 30 Hz used for Passthrough+;[21] four monochrome cameras for hand tracking[19] |
| Meta Quest 3 | Color passthrough | Two 4MP RGB cameras plus a depth sensor;[22] developer access at up to 1280x1280, 60 Hz[24] |
| Apple Vision Pro | Capture, tracking, eye tracking, depth | Two main cameras, six world-facing tracking cameras, four eye-tracking cameras, TrueDepth camera, LiDAR Scanner[3] |
| Azure Kinect | Depth | 1-megapixel AMCW time-of-flight chip, 3.5 µm global shutter pixels[29] |
| OmniVision OG0TB | Eye and face tracking | 400x400 global shutter, 2.2 µm pixels, under 7.2 mW at 30 fps[26] |
| Prophesee GenX320 | Event-based eye tracking | 320x320 event sensor, 6.3 µm pixels, as low as 36 µW[28] |
Research
Sensor research aimed at AR and VR includes work on power consumption, dynamic range and processing inside the sensor. Reality Labs and Brillnics reported at the 2023 International Image Sensor Workshop an improved global shutter stacked digital pixel sensor with 512x512 pixels of 4.6 µm, an ultra-high dynamic range of 127 dB and power consumption of 5.8 mW, with voltage regulators integrated into the same 4 mm x 4 mm die; the authors described the chip as suited to battery-powered, always-on mobile computer vision.[4] Reporting on Michael Abrash's IEDM 2021 talk, Road to VR wrote that Meta's prototype digital pixel sensor captures light values at three different light levels at once, stores the result in memory in each pixel, and draws 5 mW at 30 frames per second, which Abrash put at just under 25 percent of a typical sensor's draw. Abrash argued that this wide dynamic range is needed for future AR glasses that must work both in dim rooms and in sunlight.[32]
Event cameras remain an active research area for low-latency, high-speed and high dynamic range vision, including feature tracking, optical flow and 3D reconstruction, as surveyed by Gallego and colleagues.[11]
See also
References
- ↑ 1.0 1.1 1.2 1.3 1.4 "CCD vs CMOS". Teledyne Vision Solutions. Teledyne. https://www.teledynevisionsolutions.com/learn/learning-center/imaging-fundamentals/ccd-vs-cmos/. Retrieved 2026-10-11.
- ↑ 2.0 2.1 2.2 2.3 2.4 "CMOS Sensors Enable Phone Cameras, HD Video". NASA Spinoff 2017. NASA. https://spinoff.nasa.gov/Spinoff2017/cg_1.html. Retrieved 2026-10-11.
- ↑ 3.0 3.1 "Apple Vision Pro - Technical Specifications". Apple. Apple Inc.. https://www.apple.com/apple-vision-pro/specs/. Retrieved 2026-10-11.
- ↑ 4.0 4.1 4.2 4.3 4.4 Rimon Ikeno, Kazuya Mori, Masayuki Uno, Ken Miyauchi, Toshiyuki Isozaki, Hirofumi Abe, Masato Nagamatsu, Isao Takayanagi, Junichi Nakamura, Shou-Gwo Wuu, Lyle Bainbridge, Andrew Berkovich, Song Chen, Ramakrishna Chilukuri, Wei Gao, Tsung-Hsun Tsai, Chiao Liu (2023). "Evolution of a 4.6 um, 512x512, ultra-low power stacked digital pixel sensor for performance and power efficiency improvement". Proceedings of the 2023 International Image Sensor Workshop (IISW). International Image Sensor Society. https://imagesensors.org/Past%20Workshops/2023%20Workshop/2023%20Papers/P44.pdf. Retrieved 2026-10-11.
- ↑ Eric R. Fossum, Donald B. Hondongwa (2014-05). "A Review of the Pinned Photodiode for CCD and CMOS Image Sensors". IEEE Journal of the Electron Devices Society, vol. 2, no. 3, pp. 33-43. IEEE. doi:10.1109/JEDS.2014.2306412. https://doi.org/10.1109/JEDS.2014.2306412. Retrieved 2026-10-11.
- ↑ 6.0 6.1 E. R. Fossum (1997-10). "CMOS image sensors: electronic camera-on-a-chip". IEEE Transactions on Electron Devices, vol. 44, no. 10, pp. 1689-1698. IEEE. doi:10.1109/16.628824. https://doi.org/10.1109/16.628824. Retrieved 2026-10-11.
- ↑ 7.0 7.1 Bryce E. Bayer (1976-07-20). "US3971065A - Color imaging array". Google Patents. Eastman Kodak Co. https://patents.google.com/patent/US3971065A/en. Retrieved 2026-10-11.
- ↑ 8.0 8.1 8.2 8.3 8.4 "Sony Develops "Exmor RS," the World's First Stacked CMOS Image Sensor". Sony Global News Releases. Sony Corporation. 2012-08-20. https://www.sony.net/SonyInfo/News/Press/201208/12-107E/index.html. Retrieved 2026-10-11.
- ↑ 9.0 9.1 9.2 "Sony develops back-illuminated CMOS image sensor, realizing high picture quality, nearly twofold sensitivity and low noise". Sony Global News Releases. Sony Corporation. 2008-06-11. https://www.sony.net/SonyInfo/News/Press/200806/08-069E/index.html. Retrieved 2026-10-11.
- ↑ 10.0 10.1 "Rolling vs Global Shutter". Teledyne Vision Solutions. Teledyne. https://www.teledynevisionsolutions.com/learn/learning-center/imaging-fundamentals/rolling-vs-global-shutter/. Retrieved 2026-10-11.
- ↑ 11.0 11.1 11.2 Guillermo Gallego, Tobi Delbruck, Garrick Orchard, Chiara Bartolozzi, Brian Taba, Andrea Censi, Stefan Leutenegger, Andrew J. Davison, Joerg Conradt, Kostas Daniilidis, Davide Scaramuzza (2022). "Event-based Vision: A Survey". IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 44, no. 1, pp. 154-180. IEEE. doi:10.1109/TPAMI.2020.3008413. https://doi.org/10.1109/TPAMI.2020.3008413. Retrieved 2026-10-11.
- ↑ 12.0 12.1 Patrick Lichtsteiner, Christoph Posch, Tobi Delbruck (2008). "A 128x128 120 dB 15 us Latency Asynchronous Temporal Contrast Vision Sensor". IEEE Journal of Solid-State Circuits, vol. 43, no. 2, pp. 566-576. IEEE. doi:10.1109/JSSC.2007.914337. https://doi.org/10.1109/JSSC.2007.914337. Retrieved 2026-10-11.
- ↑ 13.0 13.1 13.2 13.3 "Milestones:Charge-Coupled Device, 1969". Engineering and Technology History Wiki. IEEE History Center. https://ethw.org/Milestones:Charge-Coupled_Device,_1969. Retrieved 2026-10-11.
- ↑ 14.0 14.1 W. S. Boyle, G. E. Smith (1970-04). "Charge Coupled Semiconductor Devices". Bell System Technical Journal, vol. 49, no. 4, pp. 587-593. doi:10.1002/j.1538-7305.1970.tb01790.x. https://doi.org/10.1002/j.1538-7305.1970.tb01790.x. Retrieved 2026-10-11.
- ↑ 15.0 15.1 "Charge-Coupled Device". Engineering and Technology History Wiki. IEEE History Center. https://ethw.org/Charge-Coupled_Device. Retrieved 2026-10-11.
- ↑ Eric R. Fossum. "Publications of Dr. Eric R. Fossum". ericfossum.com. https://ericfossum.com/Publications/Favorite%20Papers%20Table.htm. Retrieved 2026-10-11.
- ↑ 17.0 17.1 Julianne Pepitone (2018-07-02). "Chip Hall of Fame: Photobit PB-100". IEEE Spectrum. https://spectrum.ieee.org/chip-hall-of-fame-photobit-pb100. Retrieved 2026-10-11.
- ↑ Joel Hesch, Anna Kozminski, Oskar Linde (2019-08-22). "Powered by AI: Oculus Insight". Meta AI Blog. Meta. https://ai.meta.com/blog/powered-by-ai-oculus-insight/. Retrieved 2026-10-11.
- ↑ 19.0 19.1 Shangchen Han, Beibei Liu, Tsz Ho Yu, Randi Cabezas, Peizhao Zhang, Peter Vajda, Eldad Isaac, Robert Wang (2019-09-25). "Using deep neural networks for accurate hand-tracking on Oculus Quest". Meta AI Blog. Meta. https://ai.meta.com/blog/hand-tracking-deep-neural-networks/. Retrieved 2026-10-11.
- ↑ Shangchen Han, Beibei Liu, Randi Cabezas, Christopher D. Twigg, Peizhao Zhang, Jeff Petkau, Tsz-Ho Yu, Chun-Jung Tai, Muzaffer Akbay, Zheng Wang, Asaf Nitzan, Gang Dong, Yuting Ye, Lingling Tao, Chengde Wan, Robert Wang (2020-08). "MEgATrack: Monochrome Egocentric Articulated Hand-Tracking for Virtual Reality". ACM Transactions on Graphics, vol. 39, no. 4. ACM. doi:10.1145/3386569.3392452. https://doi.org/10.1145/3386569.3392452. Retrieved 2026-10-11.
- ↑ 21.0 21.1 Gaurav Chaurasia, Arthur Nieuwoudt, Alexandru-Eugen Ichim, Richard Szeliski, Alexander Sorkine-Hornung (2020-05). "Passthrough+: Real-time Stereoscopic View Synthesis for Mobile Mixed Reality". Proceedings of the ACM on Computer Graphics and Interactive Techniques, vol. 3, no. 1, article 7. ACM. doi:10.1145/3384540. https://doi.org/10.1145/3384540. Retrieved 2026-10-11.
- ↑ 22.0 22.1 "Meta Quest 3 se lanza este año + precios más bajos y mejoras para Quest 2". Meta Newsroom (Spanish). Meta. 2023-06-01. https://about.fb.com/es/news/2023/06/meta-quest-3-se-lanza-este-ano-precios-mas-bajos-y-mejoras-para-quest-2/. Retrieved 2026-10-11.
- ↑ Harry Baker (2023-08-15). "Meta Quest 3: Price, Specs, Features, Release & Everything We Know So Far". UploadVR. https://www.uploadvr.com/meta-quest-3-everything-we-know/. Retrieved 2026-10-11.
- ↑ 24.0 24.1 "Passthrough Camera API Overview". Meta Horizon OS Developers. Meta. https://developers.meta.com/horizon/documentation/unity/unity-pca-overview/. Retrieved 2026-10-11.
- ↑ 25.0 25.1 "Introducing Apple Vision Pro: Apple's first spatial computer". Apple Newsroom. Apple Inc.. 2023-06-05. https://www.apple.com/newsroom/2023/06/introducing-apple-vision-pro/. Retrieved 2026-10-11.
- ↑ 26.0 26.1 "OMNIVISION Announces World's Smallest Global Shutter Image Sensor for AR/VR/MR and Metaverse". OMNIVISION. 2022-08-24. https://www.ovt.com/press-releases/omnivision-announces-worlds-smallest-global-shutter-image-sensor-for-ar-vr-mr-and-metaverse/. Retrieved 2026-10-11.
- ↑ Anastasios N. Angelopoulos, Julien N. P. Martel, Amit P. Kohli, Jorg Conradt, Gordon Wetzstein (2021-05). "Event-Based Near-Eye Gaze Tracking Beyond 10,000 Hz". IEEE Transactions on Visualization and Computer Graphics, vol. 27, no. 5, pp. 2577-2586. IEEE. doi:10.1109/TVCG.2021.3067784. https://doi.org/10.1109/TVCG.2021.3067784. Retrieved 2026-10-11.
- ↑ 28.0 28.1 "Prophesee Launches GenX320 Metavision Sensor". Prophesee. 2023-10-16. https://www.prophesee.ai/2023/10/16/prophesee-launches-genx320/. Retrieved 2026-10-11.
- ↑ 29.0 29.1 "Azure Kinect DK depth camera". Microsoft Learn. Microsoft. 2019-06-26. https://learn.microsoft.com/en-us/previous-versions/azure/kinect-dk/depth-camera. Retrieved 2026-10-11.
- ↑ "Depth camera whitepaper - ISSCC 2018". Microsoft Learn. Microsoft. 2018-07-05. https://learn.microsoft.com/en-us/windows/mixed-reality/out-of-scope/isscc-2018. Retrieved 2026-10-11.
- ↑ Ren Ng, Marc Levoy, Mathieu Bredif, Gene Duval, Mark Horowitz, Pat Hanrahan (2005-04). "Light Field Photography with a Hand-Held Plenoptic Camera". Stanford University Computer Science Tech Report CSTR 2005-02. https://graphics.stanford.edu/papers/lfcamera/. Retrieved 2026-10-11.
- ↑ Ben Lang (2022-05-02). "Reality Labs Chief Scientist Outlines a New Compute Architecture for True AR Glasses". Road to VR. https://www.roadtovr.com/michael-abrash-iedm-2021-compute-architecture-for-ar-glasses/2. Retrieved 2026-10-11.