2D face recognition technology has come to the bottleneck. In the past two years, 3D face recognition technology has emerged. At present, the more common 3D sensing technologies have the following four types:
1. Stereo Vision: The image is captured by two camera modules, and the distance between objects is obtained by triangulation. It is the only technology that requires only RGB camera modules instead of IR (infrared) modules. Since image processing is required, an additional image computing chip is usually required, so some chip manufacturers will promote this technology.
2, Structured Light (Structured Light): The principle is to produce light streaks on the target, and then calculate the shape and distance through the changes in the light pattern that is played out, which is more common in industrial testing and research purposes. With the development of IR, Structured Light technology can also emit light through IR, so the basic components include IR emitter, IR camera module and RGB camera module.
3. Light Coding: It was used by Microsoft in the first generation Kinect somatosensory camera. The principle is that the IR laser will pass through the grating after the laser is emitted, and the light will be evenly distributed in the measurement space, and then recorded by the IR camera. Laser speckles in space, IR transmitters, IR camera modules and RGB camera modules are required on the device.
4, Time of Flight (TOF): 3DV Systems for Microsoft's acquisition, is also the second generation of Kinect technology. The principle is to obtain the time at which each point in the space reaches the observation point through the IR laser emission, and then calculate the distance to obtain a 3D depth map. Therefore, IR emitters and receivers are required, along with RGB camera modules and photosensitive components or sensing arrays.
Both Stereo Vision and Structured Light require image analysis operations, but Stereo Vision's software calculations are cumbersome, not suitable for large multi-point sensing, and the baseline length between the light source and the lens is also lengthened, which is not suitable for 3D as a whole.
In contrast, TOF can record the time data of each observation point and then calculate it. Light Coding only needs to convert the speckles in each area to calculate the distance, and the complexity is low; however, both technologies require IR emitters and receivers. In addition, memory or even a few operating parts are required, so the cost is high. In addition, the operation principle of the two is different. TOF single-point IR only needs to record time. In theory, it is easier to analyze the speckle pattern before Light Coding. Light Coding cuts the entire screen and obtains a detailed depth map. It's easier. Overall, TOF's response speed and accuracy are the best, while Light Coding performs more evenly without the need for fine depth of field maps, and Stereo Vision costs less.