Microsoft Kinect

Microsoft Kinect was a depth-sensing camera peripheral that enabled controller-free interaction with the Xbox 360 and Xbox One through body tracking and voice recognition. Launched in November 2010 under the codename Project Natal, it became the fastest-selling consumer electronics device in history — 8 million units in the first 60 days — and its underlying computer vision research produced one of the most cited papers in the history of the field.⁵

Kinect for Xbox 360 (2010)

The original Kinect used a structured light depth sensor based on technology from PrimeSense, an Israeli startup whose patents also underlaid the Occipital Structure Sensor and the original iPad Pro face sensor.³ The sensor projected an infrared dot pattern into the scene and a co-located infrared camera read the distortion of that pattern to compute per-pixel depth at 320×240 resolution, 30 frames per second.

Combined with a standard RGB camera and a four-microphone array, the Kinect could simultaneously track up to six people's skeleton poses, perform speech recognition independent of background noise, and provide depth-keyed background removal. The device required no calibration and no wearable sensors — users stood in front of their TV and the system detected them automatically.

The launch software library leaned on gesture-driven UI navigation and whole-body games: dance games, fitness applications, and sports titles where the user's body was the controller. The system tracked a 20-joint skeleton model covering the spine, limbs, and head — sufficient for games but lacking finger-level precision.¹

The Skeletal Tracking Breakthrough

The academic underpinning of Kinect's skeleton tracking was published at CVPR 2011 by Shotton et al., in a paper that became one of the most cited computer vision publications of the decade.² The method trained a random decision forest on a synthetic dataset of millions of depth images of humans in varied poses, clothing, and body shapes. At inference, each depth pixel was classified as belonging to one of 31 body parts; these part hypotheses were then used to infer joint positions through a mean-shift clustering step.

The result ran at 200 frames per second on the Xbox 360's GPU — real-time skeleton tracking from a single depth image, with no per-user calibration and robust to occlusion and body variation. This approach — large synthetic training sets, random forests over pixel features, body part classification followed by joint regression — became a template for subsequent body tracking systems and influenced the design of later ML-based tracking pipelines.

Kinect for Windows and Research Applications

A Kinect for Windows variant (2012) added a "near mode" depth range (0.4m vs. 0.8m minimum), enabling closer interaction and desktop-scale scanning. The Kinect SDK opened the depth and skeleton data to developers, generating an explosion of creative and research applications: room-scale 3D scanning (KinectFusion), interactive installations, gesture-controlled interfaces, robotic perception, and rehabilitation tools. Kinect became the de facto depth sensing platform for computer vision and HCI research from 2010 to approximately 2015.

Kinect 2 (2013)

The second-generation Kinect shipped with Xbox One in 2013, replacing structured light with Time-of-Flight depth sensing. The ToF sensor measured the time for emitted infrared pulses to return from each point in the scene, producing a 512×424 depth image at 30 FPS with substantially improved accuracy and a wider field of view than Kinect 1.⁴ The skeleton model was extended to 25 joints with improved accuracy for seated users; the microphone array was improved; and the colour camera was upgraded to 1080p.

The new depth principle also enabled better performance in bright environments — structured light is degraded by sunlight, while ToF is more robust to ambient IR — making Kinect 2 more viable for windowed rooms.

Azure Kinect (2019) and Discontinuation

Azure Kinect DK (2019) was Microsoft's developer-focused successor: a 1024×1024 ToF depth sensor, a 12MP colour camera, a spatial microphone array, and an IMU, packaged for enterprise and research use rather than consumer gaming.⁴ The Azure Kinect was positioned as an AI perception module for industrial applications — people counting, ergonomics monitoring, warehouse automation — integrating with Azure cloud cognitive services.

Microsoft discontinued the original Kinect for Xbox in 2017, and the Azure Kinect in 2022, citing the maturation of software-based pose estimation (see Body & Skeleton Tracking) that reduced the dependency on dedicated depth hardware for many use cases the Kinect had originally required.

Despite its discontinuation, Kinect's influence endures in every subsequent body tracking system. The random decision forest approach it pioneered continues in adapted form; the depth-from-structured-light and ToF technologies it popularised are found in iPhone LiDAR, HoloLens sensors, and Apple Vision Pro; and the developer ecosystem it created seeded an entire generation of spatial computing practitioners.

See also: Hardware · Body & Skeleton Tracking · Structure Sensor · Scene Reconstruction · Microsoft HoloLens · Depth Sensing

References

1. Kinect for Xbox 360 Launches Today — Microsoft News, 2010(accessed May 1, 2026)
2. Real-Time Human Pose Recognition in Parts from Single Depth Images — Shotton et al., CVPR 2011(accessed May 1, 2026)
3. Apple Acquires PrimeSense — TechCrunch, 2013(accessed May 1, 2026)
4. Azure Kinect DK hardware specification — Microsoft(accessed May 1, 2026)
5. Fastest Selling Consumer Electronics Device — Guinness World Records(accessed May 1, 2026)