What is Embedded Vision
In fact, embedded vision is the same as computer vision. As the title suggests, the only difference is the hardware. Usually, it is a large advanced computer chip mounted on a tiny circuit board. These chips are powerful and work well for a variety of vision tasks.
Embedded vision is used for two main reasons. An alternative to computers when advanced algorithms cannot be efficiently executed on ordinary computers.
The first reason is the most obvious. A single-board computer like Odroid-UX4 can be used, why use an expensive computer every time? Because there are fewer parts, a smaller space is required and less downtime is required. Finally, it works well in harsh environments.
Odroid-UX4
The second reason for using embedded vision has to do with the computer hardware itself. Computers are optimized for general-purpose arithmetic and logical operations. For specific operations like the Fourier transform's butterfly memory addressing scheme it's less important. In this case, an embedded system with custom electronics will outperform a computer. Of course, before starting to build your own electronics, it's advisable to do a research to see if this method will work.
Not only can computers be replaced with embedded hardware, but so can cameras. This can further reduce the unit price. Note that not every tiny embedded camera has great optics. Sometimes this can be a highlight.
Embedded camera for Raspberry Pi
Camera vendors also sometimes sell so-called embedded vision systems or smart cameras. The camera and processor are usually in the same space. There are some caveats here. Beyond obsolete processors, there's always friction between manufacturers' reluctance to reveal the secrets of their hardware and software, and programmers' desire to configure an optimal design. The situation is even worse when the embedded vision camera has only one library. Needless to say, this was what led to the disappointment.
Is embedded vision always a better choice than "normal" computer vision? It depends on the problem at hand. Still, regular computers, especially those with graphics cards, outperform all standard embedded systems. Therefore, if you want to run some large-scale deep learning programs or training tasks, computers are still a better choice. However, even these applications are increasingly being replaced by embedded systems, think self-driving cars.
Another reason to stick with a regular computer might be development. For the common computer environment, there are a large number of software tools available. Especially when writing applications that involve a lot of trial and error, good debugging tools can go a long way. At a later stage, applications can still be ported to embedded systems.
At a later stage, applications can still be ported to embedded systems.