Hand-tracking
Solutions
A software-and-hardware integrated hand-tracking solution for AI glasses, AR, and VR smart glasses and other wearable devices. It combines JIIOV's JVH imaging module, LED array, flexible printed circuit board (FPC), and proprietary deep learning algorithms into a compact, low-power form factor that is easy to integrate, supporting one or two imaging modules per pair of glasses.
JVH offers two core capabilities:
- Hand Keypoint Tracking:
Detects the user's hand position and continuously estimates overall hand pose and joint pose in world coordinates. - Gesture Interaction:
Parses temporal tracking output to recognize semantic user gestures such as confirm, tap, drag, and handwriting input.
Features
- Dual-hand detection and tracking with left-right hand differentiation
- Outputs 3D information of 26 hand joint points for both hands
- Supports 3D hand mesh
- Supports hand depth perception
- Recognizes various static and dynamic gestures, with support for custom gestures
Typical Technical Specifications
| Category | Item | Specification | Notes |
|---|---|---|---|
| Checkout | Detect left and right hands | Normal>99%, Occlusion>95%, False detection<0.1% | Confidence output for enhanced rendering and interaction control. |
| gesture | Recognition rate | Palm>99%, Pinch>95%, Fist>95% | Except for the palm, other gestures are easy to miss at certain angles. Strategy optimization can be used for normal interaction experience. |
| Key Points | Average keypoint error | <1.5cm | Absolute error of bone key points |
| Hand shape accuracy | <0.8cm | Relative accuracy, the relative positions of different joints of a single hand | |
| Static Floating/Dynamic Floating | <0.3cm | Stability, jitter of algorithm results when the hand is not moving | |
| Speed Latency (XR2 Platform) | Algorithm time | 9ms two-hand tracking | According to the output frame rate, it can reach 30FPS |
| Algorithm delay | 23ms @60FPS | The smoothing strategy brings <1 frame delay, and the superposition algorithm takes time | |
| Resource usage (XR2 platform) | Memory | Within 100M | |
| CPU | <20% | ||
| DSP/NPU | <10% | ||
| Power consumption | 200mW | Reference value, can be optimized according to the whole system status |
Key Advantages
High performance in complex scenarios
- Excellent recognition efficiency and accuracy in complex finger interactions (such as pinching or twisting) and unconventional gestures
High R&D investment
- Ten million level test database
- Multiple core patents in gesture recognition
Industry-leading gesture recognition algorithm capabilities
- Best performance at the 7th HANDS workshop challenge – AssemblyHands 2023
- Best performance in the Multiview Egocentric Hand Tracking Challenge at the 8th HANDS workshop 2024
- First Place performance at DexHAND workshop challenge - SHOW3D track 2026
- Multiple research papers published at top-tier international conferences (CVPR, ECCV, etc.)
Solution Function Demonstration
Supports multiple static and dynamic gesture recognition and custom gestures;
Accurate and smooth bare-hand interaction can be achieved in complex situations (occlusion, overlap, etc.)
