High speed, close, low latency
Editor’s note: This article comes from WeChat public account “ Quantum Bit “(ID: QbitAI), author thirteen. p>
This black technology makes AE effects come true in seconds. p>
Not long ago, a plugin called Lockdown from AE became popular: no matter how uneven you move, the pattern will fit seamlessly. p>
p>
Actually, this magic special effect has come into reality. p>
This black technology is actually the projection we know. But unlike in the past, this time is a dynamic projection. p>
This high-speed projector is called DynaFlash. p>
The projected picture does not need to be flat. If the picture is distorted or blocked, the image can be completely printed. p>
p>
If you want a better projection experience, the images you cast must be “fast, accurate, and cruel”. p>
It requires high frame rate and low latency. p>
This high-speed dynamic projector can project 8-bit images with a delay of 3ms, up to 1000 fps. p>
This topic has also become popular on Hacker News, with a heat of 1000+ points. p>
p>
Many netizens exclaimed that the technology is “awesome”!! p>
I expect companies like Disney to be very interested in this technology. p>
I have been working in real-time projection for many years. Delay is the killer of this job. Any delay greater than 1000 / 90ms will have a special effect on the effect. This technology allows fast movement. p>
Cool Black Technology h2>
How amazing is this black technology? p>
The image is projected onto a piece of paper. No matter how shake it is, the image will change synchronously with the deformation of the paper. p>
As if the image was pasted on paper. p>
In order to demonstrate the high frame rate and low latency of this technology, the research team also slowed down the video. p>
The dynamic projection under slow motion also achieves a perfect fit. p>
p>
Of course, rotation doesn’t matter. p>
p>
What’s more amazing is that the images can be switched freely with different projection screens! p>
What is it? h2>
The black technology shown in the video actually uses infrared ink to draw marks on the target. p>
p>
This mark is invisible to the human eye. p>
This is the Deformable Dot Cluster Marker (DDCM) proposed by researchers. p>
DDCM has three advantages: p> address>
1, even in the case of strong deformationIt can also be detected; 2. It can achieve robust tracking even when there is external or self-occlusion; 3. It allows millisecond-level calculation speed. p>
The following figure shows an example of DDCM. p>
p>
△ An example of DDCM. Point clusters are distributed in an array structure. p>
DDCM consists of four types of point clusters, which are distributed in an array structure. As shown below. p>
p>
Four types of delta point clusters. m represents the number of points. Each point cluster contains 1-4 points. p>
The points in each point cluster are topologically invariant, that is, they will not change under any deformation. p>
The number of
points is randomly and exclusively distributed. Recognition starts from the four corners of the p x p point cluster (recognition window). p>
In order to enhance the robustness of frame-by-frame tracking, the number of points in adjacent point clusters will not be the same. p>
p>
△ No matter which corner is used to identify the window, there are no sequences that overlap each other. p>
The following figure shows the processing flow. p>
p>
△ Processing flow p>
It consists of two steps: detection and tracking. p>
The detection process involves identifying the clusters of points from a single frame. p>
p>
p>
△ Detection process p>
The detection process is roughly as follows: p> address>
a) extract key points from the binarized image; b) consider the key points shown in a) located in the same grid and adjacent to each other as a cluster of points, red, green, blue, and yellow Representing m = 1, 2, 3, and 4 point clusters respectively; c) Calculating the Delauney triangulation of the extracted point clusters; d) Obtaining connectivity between point clusters based on breadth-first search; e) Finally, according to the pre-prepared The hash table identifies the ID of each point cluster according to the point number, and the red number indicates the ID of the identified point cluster. p>
The tracking process refers to independent, frame-by-frame tracking of each point cluster through an image sequence. It mainly includes the following processes: p>
a) Update the location of the tracked point cluster; p>
p>
△ Update the tracked point cluster position according to ROI, which is indicated by a red square. p>
b) Track the recovery of lost point clusters; p>
p>
△ DDCM-based array array structure tracking recovery. p>
c) False-positive tracking detection; d) interpolation of missing point clusters. p>
More ways to open h2>
Think black technology can only do this? p>
The dynamic projection shown above only stays in the black and white color stage. The next way to open is even more cool. p>
With the countdown of “3, 2, 1”, the clothes suddenly appeared a dynamic projection with bright colors and complex textures. It is still you to pull, I am like a shadow. p>
p>
Of course, there is a one-second drag wind operation. p>
The texture of the texture can also be changed, and with different actions and angles, the light and other factors will also change. p>
p>
The naked eye 3D is no problem! p>
p>