Global shutter

In photography, global shutter means every pixel is captured at the same time. In an electronic image sensor this means that all pixels start and stop their exposure together, and the exposure is ended by moving each pixel's signal into a memory inside the pixel, from which it is read out afterwards. The result is a snapshot of a single point in time.

In the traditional rolling shutter, every pixel is not captured at the same time. Pixels from different rows are captured at different times. Fast-moving objects therefore appear skewed or distorted in a rolling shutter image, while a global shutter records them without that distortion.

In virtual reality and augmented reality hardware, global shutter sensors are used in cameras that track the headset, the surroundings, the eyes and the face. Oculus used a global shutter sensor in the positional tracking camera of the Oculus Rift DK2, and image sensor makers such as OmniVision and Samsung sell small global shutter sensors aimed at eye tracking, face tracking and SLAM cameras in XR headsets and glasses.

• This term is applicable to Cameras

Reviewed 27 September 2026. Checked every cited source against the new text: sensor definitions and trade-offs (Axis, Teledyne, Sony), pixel figures in the two MDPI Sensors papers, the Rift DK2, T265 and Project Aria camera specs, the OmniVision and Samsung sensor announcements, and the three tracking papers. About review dates.

How it works

Rolling shutter readout

Conventional CMOS image sensors use a rolling shutter, in which each row of pixels is exposed in sequence and there is a slight time difference in the readout of each row. Sony describes a rolling shutter as one "in which the sensor processes photon-converted electrons for each line of an array and reads them out one by one", and a global shutter as one "that stores up the electrons and reads out as a complete image". In a rolling shutter sensor each row has a line time, and the frame time adds up across all rows. Teledyne gives the example of a sensor with 2048 rows and a line time of 10 microseconds, so the bottom row is read out 2.048 ms after the top row; Teledyne notes that this delay adds up across large sensors and should be considered when imaging very fast subjects.

Global shutter pixels

A global shutter CMOS pixel needs somewhere to hold its signal between the end of the exposure and the moment its row is read. In Sony's Pregius sensors, the light sensed by the photodiodes of all pixels at the same time is transferred to memory sections in each pixel before it is processed. Mizuno and colleagues divide global shutter CMOS pixels into two families. In charge-domain pixels, the charge generated in the photodiode is moved to an in-pixel memory node. In voltage-domain pixels, the signal is amplified in the pixel and stored as a voltage on capacitors. The charge-domain approach can perform true correlated double sampling with a single extra memory node, while the voltage-domain approach needs at least two switching transistors and capacitors, so the authors consider charge-domain pixels better for small pixel sizes and low noise.

The in-pixel memory brings its own problems. Light that reaches the memory node instead of the photodiode adds charge to the stored signal after the exposure has ended. Yokoyama and colleagues call this parasitic light sensitivity (PLS) and note that "an increase of PLS leads to deterioration of image quality". Suppressing dark current in the memory node is a second requirement for a high-performance global shutter sensor. Published pixel designs report these figures directly:

Pixel design	Pixel pitch	Reported 1/PLS	Reported quantum efficiency	Memory-node dark current
Yokoyama et al. (2018), storage-gate memory node	2.8 micrometres	7,700	62% at 530 nm	9.5 e-/s at 60 degrees C
Mizuno et al. (2020), charge-domain pixel	2.5 micrometres	10,400 (with an f/2.8 lens)	68% at 530 nm	13 e-/s at 60 degrees C

Sources: Yokoyama et al. 2018; Mizuno et al. 2020.

Trade-offs

The extra circuitry has costs. Axis Communications notes that a global shutter pixel is more complex than a rolling shutter pixel, which makes the sensor more expensive, and that noise is typically higher; compensating for the noise calls for a larger sensor with larger pixels and a lens with a low f-number. Teledyne lists increased read noise, limited frame rates and longer duty cycles among the drawbacks, and adds that because a global shutter needs additional electronics on the sensor, it is typically most suitable for front-illuminated sensors, which it describes as typically limited to 70-80% quantum efficiency. Teledyne also describes global shutter operation as typical of CCD and EMCCD cameras.

Some newer designs stack the circuitry under the pixels. Sony's Pregius S technology combines its global shutter with a back-illuminated pixel structure and a stacked structure that places the pixel section over the signal processing section, which Sony says frees circuit area for additional functions. OmniVision's OG0TB, announced in August 2022, is a three-layer stacked back-side illuminated global shutter sensor.

Rolling shutter artifacts

Because the rows of a rolling shutter frame are exposed at slightly different times, several kinds of distortion appear that a global shutter avoids:

• Skew: objects moving quickly in a straight line through the image, such as cars or boxes on a conveyor belt, look tilted.
• Distortion of moving objects: the time offset between rows changes the shape of fast-moving objects, known as the rolling shutter effect. Samsung calls this the "jello effect".
• Partial exposure: a short flash lights only the rows that are exposing at that instant, producing a bright band across the image.

Sony's Pregius documentation describes the global shutter as able to capture a high-speed moving image "without focal plane distortion".

Applications in VR and AR

Tracking cameras

Camera-based tracking in VR and AR estimates the pose of the camera from the positions of features in its images. A rolling shutter complicates this, because the rows of each image were recorded from slightly different positions when the camera is moving. Lee and Yoon note that relative pose estimation "has been studied for conventional global shutter cameras for decades", and that rolling shutter cameras have become widely used because of their low cost, but that line-by-line capture makes estimating their relative pose harder than for a global shutter camera. Li and Mourikis studied the same problem for visual-inertial navigation with an inertial measurement unit and a rolling shutter camera. They noted that the usual approximate, low-dimensional models of camera motion can lose accuracy when the motion contains significant accelerations, and instead used the inertial measurements to avoid assumptions about the trajectory, in an extended Kalman filter that they demonstrated on smartphones.

For the Oculus Rift DK2, Oculus used a global shutter image sensor in the positional tracking camera and synchronized it with the infrared emitters on the headset. The emitters were modulated through a set of patterns in Oculus's early implementation of Constellation tracking, according to the hardware notes Oculus published when it open-sourced the DK2 design in 2017.

Intel's Intel RealSense T265 tracking camera runs visual SLAM on the device with two fisheye cameras, each with a circular field of view of about 165 degrees captured by "~800 pixel diameter monochrome global shutter cameras", together with an IMU and an Intel Movidius Myriad 2 vision processing unit; it reports camera pose at 200 Hz. The RealSense documentation notes that most AR and VR headsets use their own versions of visual-inertial odometry to track the full six degrees of freedom.

Meta's Project Aria research glasses show how the two shutter types can be divided among the cameras of one device. The first-generation glasses, described by Meta Reality Labs researchers in 2023, use global shutter sensors for their two monochrome scene (SLAM) cameras and two eye tracking cameras, and a rolling shutter sensor for the RGB camera:

Aria Gen 1 camera	Count	Shutter	Maximum resolution	Field of view	Maximum frame rate
Mono scene (SLAM)	2	Global	640 x 480	150 degrees H x 120 degrees V	30 fps
Eye tracking	2	Global	640 x 480	64 degrees H x 48 degrees V	90 fps
RGB	1	Rolling	2880 x 2880	110 degrees H x 110 degrees V	30 fps

For Aria Gen 2, Meta doubled the number of computer vision cameras to four and stated that they use a global shutter sensor with a dynamic range of 120 dB, compared with 70 dB in the first generation.

Eye, face and gesture tracking sensors

Global shutter sensors for inward-facing eye tracking and face tracking cameras are sold as a distinct product category. OmniVision states on its eye tracking application page that "the reduction and elimination of image artifacts is a must" for high-quality, real-time experiences. Examples announced for XR devices include:

Sensor	Maker	Announced	Resolution	Pixel size	Stated XR uses
OG0TB	OmniVision	August 2022	400 x 400	2.2 micrometres	Eye and face tracking in AR/VR/MR devices; 1.64 x 1.64 mm package, under 7.2 mW at 30 fps, Nyxel technology for quantum efficiency at the 940 nm near-infrared wavelength
OG0VE	OmniVision	December 2022	640 x 480	3.0 micrometres	SLAM, gesture detection, head and eye tracking, depth and motion detection; up to 60 fps
ISOCELL Vizion 931	Samsung	December 2023	640 x 640	Not stated	Iris recognition, eye tracking, facial and gesture detection in XR headsets; 60% quantum efficiency at 850 nm; multi-drop connection of up to four cameras on one wire
OG0TC	OmniVision	July 2024	400 x 400	2.2 micrometres	Eye and face tracking in AR/VR/MR headsets and glasses; first OmniVision AR/VR global shutter sensor with its DCG high dynamic range technology

OmniVision said the OG0TC's small package matters for headset design because several cameras are needed to track all parts of the face, including the eyes, brows and lips.

Research using rolling shutter for tracking

Not all tracking research treats the rolling shutter as a flaw. In a 2018 CVPR paper, Akash Bapat, True Price and Jan-Michael Frahm of the University of North Carolina at Chapel Hill used the row-by-row timing of rolling shutter cameras, together with lens radial distortion, to estimate a separate 6-DoF pose for each image row. Using clusters of four or six commodity rolling shutter cameras, their system tracked at up to 86.4 kHz. The authors argued that high frame-rate global shutter cameras are not a solution to kHz tracking, because the maximum exposure time falls to the inverse of the frame rate, which makes it impractical to collect enough light indoors with the small cameras a headset can carry.

See also

• Rolling shutter
• Inside-out tracking
• Visual-inertial odometry
• Eye tracking
• Hand tracking
• Camera