How close are virtual digital people to real people? Recently, virtual people have developed explosively, and different industries such as banking, media, and beauty have launched their own virtual people. So what exactly is a virtual digital human? According to the “2020 White Paper on the Development of Virtual Digital Humans” issued by the China Artificial Intelligence Industry Development Alliance, a virtual digital human is a virtual character with a digital appearance, which should have human appearance, human behavior, and human thinking.
The virtual digital human image is divided into two categories: 2D and 3D, and the appearance style is divided into cartoon, realistic, realistic and other types. For example, Ling, Liu Yexi, etc. are typical 3D hyper-realistic virtual digital people. Hyperrealism means that the appearance of the characters is highly simulated and lifelike. This kind of virtual human needs more than 10,000 faces, and can withstand 360-degree shooting without dead ends with high precision.
The facial material is not only very close to the texture of real skin, but can also be automatically optimized according to the distance of the camera, and the skin, facial features, hair, and limbs are almost real.
Are you curious how this 3D hyper-realistic virtual digital human is created? To put it simply, the production of a 3D virtual digital human needs to go through three steps: image generation, animation generation, and voice generation. Image generation determines the appearance of the virtual human, animation generation enables the virtual human to move flexibly, and speech generation allows the virtual human to speak, express and interact.
1. Image generation
The most important part of image generation is modeling. Common modeling methods include manual modeling, scanning modeling and AI modeling. With the development of science and technology, more efficient scanning modeling and Al modeling technology have gradually become the mainstream way of character modeling. The CCTV network digital virtual editor Xiao C adopts the camera array scanning method for modeling. In the scanning studio integrating more than 100 cameras, a large amount of facial and body data is collected into the system with only a short acquisition time. , obtain the most original three-dimensional character model data through algorithm processing.
In addition, in order to collect more accurate images, more advanced 4D scanning technology can also be used, which can capture more subtle movements of the face and collect richer dynamic data than 3D scanning.
After the modeling is completed, to make the icy model move, a series of binding and driving are required. Bone and muscle bindings determine the naturalness and fluidity of the model’s subsequent body movements and facial expressions. At present, there are two mainstream methods: skeleton binding and hybrid deformation binding, and the driver is divided into human driver and intelligent driver. Human driver refers to the capture technology to collect the movement and facial expression data of real actors and then migrate and synthesize these data into virtual figures. On people, optical capture and inertial capture are common motion capture methods, but both require wearing motion capture equipment, and the threshold for use is relatively high.
In recent years, the capture technology based on computer vision has developed rapidly. The facial expression capture of hyper-realistic virtual digital human can also be collected through a depth-of-field camera to capture the 3D point cloud image of the real person’s face, and then the facial movements and expressions can be transferred to the virtual human in real time. , compared with the need to wear motion capture equipment and rent a motion capture shed, only a mobile phone is required to drive the virtual human, which makes the operation process more convenient.
2. Animation generation
With binding and driving, it is also necessary to generate animations through rendering. Rendering is divided into real-time rendering and offline rendering. In order to realize real-time control and real-time interaction of virtual digital humans, major rendering engines have been making breakthroughs in algorithms and improving real-time rendering efficiency. , and hope to obtain the optimal solution in the picture quality, rendering speed, and computing resources of real-time rendering.
3. Voice generation
After the image and actions are completed, the virtual human can speak. The voice of the virtual digital human can use synthetic voice or real voice. After artificial intelligence technology and continuous training, the synthetic voice will become more and more similar to the tone, rhythm and cadence of real voice. And it can correspond to the lip shape in real time, and the real voice is the voice of the person in use directly or the voice actor dubbing, and the voice of the person in the middle can be converted into the same voice through the voice changer, so that even if the person in the middle changes, the voice It will not change, which provides a certain convenience for the stability of the avatar image.
Virtual digital human is the perfect combination of technology and art. Unconsciously, more and more digital virtual human is gradually entering all aspects of society, and through different images, personalities and functions, it brings more and more opportunities to the work and life that everyone is accustomed to. A fresh, warmer experience.
This is due to the innovation and promotion of digital virtual human technology by the entire industry, as well as players’ continuous attempts to break the boundaries of technology and promote technology integration. In the virtual space full of imagination in the future, virtual digital human is also freer and more ideal. As the provider of virtual human technology in the industry, Qiwei Technology and Zengtuo are also striving to develop towards an intelligent and personalized path, and use technology to achieve innovation and empowerment for all walks of life.
About The Author
Leave a comment