The 3-D posture estimation

The estimation of posture (pose) of human body in 3-D space is far more difficult than the posture estimation in 2-D space. The 3-D posture estimation has been remarkably developing in recent years.
The 3-D posture estimation is expected to be utilized for various areas such as detection of suspicious behaviors or movements of robots or animation. However, generally, it is difficult to estimate a 3-D posture very accurately from color images. Expensive equipment such as motion capture consisting of several sensors or body suits are typically required. Such equipment is also disadvantageous for its utilization outdoors.

3-D posture in real time called “GW-Pose”

Global Walkers has developed a technology to estimate a 3-D posture in real time called “GW-Pose”.
It measures the positions of articulations of human body in 3-D space by leveraging color images obtained from a Depth camera and distance information. It can estimate a 3-D posture in real time from moving images. Within a measurable distance for the Depth camera, GW-Pose can be used anywhere including factories, sporting event venues or office environment.

Feature of GW-Pose

Accuracy

Accuracy

GW-Pose realizes far more accurate posture estimation than that of AI models using RGB values only, by using distance information obtained from the Depth camera such as Intel Realsense.

Computational Cost

Computational Cost

With a common laptop PC (equipped with GPU of NVIDIA), the estimation processing with about 10 to 100 FPS (frames per second) is available. With a small computer such as NVIDIA Jetson, the real-time processing is available.

Modular

Modular

GW-Pose is compatible with X86 or ARM processors. Concerning the Depth camera, you can use a camera such as a ToF camera other than Intel Realsense.

Real-Time 3-D output

Real-Time 3-D output

GW-Pose can be used for the output with up to 100 FPS for 3-D coordinates of 18 human body areas.
If you wish to measure the 3-D coordinates of any body areas other than the above specific 18 areas, you can customize them.

Multiple people detection

Multiple people detection

Even if several persons exist in a camera view, all of their 3-D postures can be simultaneously estimated in real time at the speed and with the accuracy equivalent to those in the case of existence of a single person in a view.

No Special set-up requires

No Special set-up requires

The necessary devices are one computer and one Depth camera only.
Since its system configuration is simple and compact, it can be loaded into a robot or drone.

Demonstration

Single Person

Outdoor

Office Working

Viewpoints