Hello everyone, I’m back again with another interesting blog.
Following the discontinuation of Kinect in October 2017, those who relied on it might have felt disappointed and perhaps a bit lost.
Today, I’m going to give you an insight into our development efforts with VisionPose. I’ll also explain how VisionPose got started – our answer to filling the void left by Kinect.
Also, we are proud to announce that our company is no exception. We are actively engaged in selling a wide range of products.
After the announcement of the suspension of Kinect sales, we immediately hustled to secure some Kinect units before they sold out!
The image below features one of our employees surrounded by significant number of Kinects.
However, relying solely on that aprroach could pose challenges in the future.
We really wanted to detect skeletons without relying on cameras with depth sensors! So, we looked at possible alternatives.
In particular, we focused on developing Skeletal Detection Systems using Deep Learning. After various considerations, trials, and experiments, we decided to make our own product because Kinect sales had been discontinued, and similar products were a bit expensive!
That is why we decided to create an alternative to Kinect. One of the reasons for this decision was that our company already had expertise in image recognition using Deep Learning.
And the name of the Skeletal Detection System, utilizing web cameras and Deep Learning, is VisionPose.
What is VisionPose?
It is a system that utilizes Deep Learning without relying on depth sensor cameras, allowing the detection of the human skeleton using only a web camera.
It can perform almost real-time bone tracking, even for complex movements such as baseball, tennis, yoga, and more.
※Compared to similar products, the mechanism for detecting neural network structures and joint connections is different.
What is the best thing about VisionPose?
Detect skeleton in real time
It supports real-time measurement and is compatible with both 2D and 3D (depth measurement) skeletal detection. Since it can extract skeletal information using only a webcam, it does not depend on depth sensors.
Capable of detecting skeletons of multiple people
It can track in real-time for not just one person but multiple people in real time, detecting the skeletal structure of each person.
Covers 30 measurement points comprehensively
It can detect joints in various parts of the body (25 points) as well as facial features (5 points). Depending on your needs, it is possible to add new measurement points through consultation.
Exceptionally user-friendly! Provided as a C# SDK without usage restrictions
VisionPose is offered as a C# SDK, making it easy to integrate into applications. In addition, there are no usage restrictions, and we offer it as a one-time purchase.
Capable of analyzing images
In addition to the real-time version, it is also possible to perform skeletal detection using still images from your collection that were captured in the past. This feature is particularly recommended for situations where real-time performance is not crucial and higher accuracy data is desired, such as with image data captured for research purposes.
Please note that coordinate detection from still images is limited to 2D coordinates only.
Supports multi-device compatibility
We aim for future compatibility on the cloud and smartphones. We are currently working on optimizing and speeding up the current model to ensure it can operate even on low-spec hardware, such as smartphones.
The only product made in Japan with Excellent Accuracy
Crafted with precision and pride, our product embodies the quality synonymous with ‘Made in Japan.’ Each component is meticulously designed and manufactured to uphold the highest standards, ensuring not only exceptional accuracy but also a durable and reliable performance.
We are ready to assist with various inquiries. Please feel free to contact us at any time.
Frequently asked questions
Merits of VisionPose
<Eliminates the following disadvantages commonly found in cameras with infrared sensors>
- Only bones from the front can be captured. If it recognizes the back, it may be misinterpreted as the front.
- Since only bones from the front can be captured, there is distortion in bone detection when switching from the front to the back (Due to the occurrence of left-right reversal).
- Recognizes only Depth Maps through infrared (without using color), making it challenging to distinguish between obstacles and people. This results in weaknesses when parts of the body are obscured or when performing actions involving holding objects.
- Difficult to recognize outdoors due to infrared rays.
- Weak against clothing that absorb infrared rays.
After repeated learning and improved accuracy, we compared skeletal detection in real-time with an existing depth sensor-equipped camera.
<User-friendly and Flexible Sales Format>
In conventional similar services utilizing deep learning, there have been limitations in specific fields or high annual fees for commercial use that are often expensive, making it less accessible and user-friendly.
With VisionPose, we aim to make it more convenient for our customers to use by offering unrestricted commercial use without specific application limitations. Additionally, our pricing model involves a one-time purchase, providing you with a simple and accessible option.
Furthermore, you will receive a product key upon purchase of the VisionPose.
After your order is completed, we will email you the product key as well as a link to download the software in 1-3 business days. For more detailed explanation, please contact us first.
What are the possible use cases?
We have been receiving inquiries for various fields such as factory operations, healthcare, and sports, for research purposes or as an alternative to Kinect. Recently, there has been an increasing number of inquiries regarding the use of motion capture for applications.
With a wide range of application fields, such as motion analysis in sports and fitness, workflow analysis and hazard detection in factories, safety surveillance in child and nursing care, motion captioning in entertainment and gaming, VisionPose is currently used by more than 400 clients now, including several major Japanese companies like Toyota Motor Corporation, NEC Solution Innovators, Ltd, Avex Management Co., Ltd., Konami amusement etc.
- Factories/Retail: Surveillance tasks, etc.
- Sports: Form checks and referee materials.
- Medical: Posture evaluation and R&D for rehabilitation and healthcare
- Embedded: Integration into smartphones and IT devices
- Entertainment: Virtual YouTuber(Vtuber), MMD production, etc.
For more information, please click here.
Renowned Figure ‘YUJI OHDOI’ experiments with VisionPose
We also had the former bassist of the Japanese band ‘The Checkers‘, YUJI OHDOI, experience a demo of VisionPose.
Due to the specifications of VisionPose, it’s difficult to tell if it’s Mr. Ohdoi as his face is obscured, but you can see a glimpse of his mischievous nature.
Thanks to Mr.Ohdoi for graciously accepting such a challenging request.
By the way, the person with messy blonde hair is our General Manager and Chief Producer.
About Annotation
What if the skeletal detection accuracy is poor?
To prepare for such situations, VisionPose has been made capable of additional learning, and we have summarized it as clearly as possible.
VisionPose is trained to recognize common movements in daily life to make it versatile for general use. Therefore, there may be a tendency for accuracy to decrease when it comes to movements that are less common in daily life.
However, our company provides a learning environment, including annotation tools, allowing for additional training to customize and further improve accuracy.
For more details, please check the link below!
Annotation tool is an optional.
Conclusion
I have compiled the current development information about VisionPose.
We plan to continue updating the latest information on platforms such as Twitter & facebook.
Stay tuned for further updates! 😊
Table of Contents
