Motion capture has an important role in all walks of life. There are two main ways we can achieve it. One is the way often used in movies and games. It records the motion trajectory by attaching markers at the joints;
The second is a depth camera, which measures the user's body through infrared. This is very familiar to everyone. The former cannot be universally available. The latter is inefficient under strong light, but has a common good quality—expensive.
The market is full of ordinary smart phones, if you can only achieve physical tracking through the camera of ordinary smart phones, then this will be a huge step forward.
In addition to being cheap, the other big advantage of the ordinary camera over the depth camera is the wide field of view that can be used for large scenes. At the same time, the infrared-based depth camera is difficult to work under outdoor direct sunlight, but the 2D camera does not have such annoyance.
Currently, scientists at the Max Planck Institute for Scientific Research have developed a system that can capture 3D motion in video in real time using only a single common 2D camera (such as the camera of an ordinary smartphone). The effect can be comparable to the effects of multiple cameras or depth cameras such as Kinect.
The use of 2D cameras to obtain 3D images is something that developers at the forefront of technology have been doing. We have mentioned many times that the simple point is to learn 3D movements through 2D images through deep learning. For example, a 3D picture of a person's appearance can be made through a photo.
The use of 2D cameras to capture 3D motions is also what scientists have been doing, and skeletal posture is one of the challenges. In previous solutions, the 3D joints were very unstable and would have serious deviations, which would not accurately match the 2D position when re-projected. Even if it is accurate, it is only in a very short period of time and it is unstable for a long time. This results in most of them only stably capturing local 3D movements, not the entire body.
VNect solves the problem of time and can stabilize the skeletal posture in a short period of time while capturing body movements.
VNect uses a Convolutional Neural Network (CNN) to abandon the previous 100-layer design, because the operation time is too long and it is not suitable for real-time operation, and it is changed to a shallower 50-layer. At the same time, a new formula makes it possible to output with high precision and operate at 30Hz or more at the same time.
In CNN's skeletal prediction, the most important 3D pose of the pelvis, the result of their proposed pose formula, can be compared with the world's most advanced offline method. The combination of 2D and 3D joint position prediction to ensure joint stability, while applying filtering steps to ensure smooth tracking over time, improve the stability of the tracking attitude.
According to the picture, a 2D heat map is developed, and then the relative positions x, y, and z of the joint are captured and become joint H. Calculating the length of the skeleton based on the length between the joints will be an important basis for predicting the movement. The usual 17 joints were collected and toes were also taken.
The system uses the average bone data. At the beginning, there will be unstable phenomenon, because there is no other data for a single camera to refer to, so the tracked object needs to provide a height data to achieve stable tracking.
At the same time, the system is completely convoluted, can operate in the absence of a compact around the target, and can predict different scenarios, which undoubtedly provides a solid basis for accurate measurements.
During the training, eight items were selected, five with the chest, two with the head and tilted downwards, one with the knee and tilted upwards to learn to maintain stability at different angles. At the same time, a variety of combinations are used, such as background, clothing (increasing clothes), obstacles (such as chairs), and the like.
The result is gratifying. In the use of low-end mobile phone cameras, the quality is comparable to that of a depth camera in terms of 3D character control, posture tracking, etc. For example, the Kinect on the Xbox (that is, the first generation and the second generation, not the latest four generations). ).
In terms of speed, the CNN calculation needs 18ms, the skeleton fitting needs 7 to 10ms, the preprocessing and filtering needs 5ms, and the maximum time is only 33ms.
In addition to the advantages introduced in the previous article, this solution also has features that depth cameras do not have, such as sitting on a stool or other limbs close to the scene objects, Kinect obvious confusion, and 2D solution is successful. However, when the legs are crossed, it is more difficult and it will be a challenge.
With VNect, any smartphone can become a lightweight, fully automatic handheld motion capture sensor, which will make many applications possible. At the same time, it is still open source, allowing more people to participate in the development.
However, it also has many shortcomings. VNect's ankle tracking is very stable, but the head accuracy is very poor. And a single 2D camera is severely deficient in depth capture, and any slight inaccuracy can lead to great differences. It is not long enough in time. Tracking can also occur in long time. In addition, occlusion is still a very serious challenge, and very fast movements are also.
2.54MM Pitch Signal Power Connector
2.54MM Pitch Signal Power Connector.Chip power connector is used in power module system. It can select the matching power + signal connector according to the need. The feature is that the number of power and signal contacts and the matching sequence can be selected arbitrarily while keeping the connector size and contact core number unchanged.
Plug (male) / socket (female) can be installed at 90 or 180 degrees. It supports mixed or independent combination of signal and power. The quantity range of power and signal is (2-16) pin and (12-128) pin respectively
Product features
High temperature resistant, glass fiber reinforced and flame retardant polyester is used as insulation material
Copper gold composite conductor with high conductivity is used, and the contact area of the conductor is plated with gold
It adopts shrapnel contact, which has the characteristics of integration, small volume, large current carrying capacity, soft plug-in, blind plug-in, self guidance and high dynamic contact reliability. This series of products can be interchanged with FCI's powerblade series and Tyco's multi-beam series
The length of power hole / signal pin can be selected in two sizes. The power rated current is 45A and the signal rated current is 2.5A
2.54MM Pitch Signal Power Connector
ShenZhen Antenk Electronics Co,Ltd , https://www.antenksocket.com