Knowledge and book sources needed to finish an embedded system design project

[AndresJeremias]

Hi everybody.

I am new to this forum and feel grateful in advance for the time you have taken to read my question.


I have taken a long term based weekend Project which implies designing and prototyping a motion capture gauntlet (http://en.wikipedia.org/wiki/Motion_capture) and would need some advice related to the new knowledge I should learn in order to finish my objective.

I am a third year student of electronics engineering and I am very curious. I found the MITopencourseware (http://ocw.mit.edu/index.htm). I think the content there and in the rest of the web could help me but as I said before, I am not very sure which new things I should learn, and here is where my question comes.


Which new theoretical and practical knowledge/subjects you think I should learn in order to finish my Project (Could you recommend any source?), considering the fact that I have a decent to basic skills or understanding in the following subjects/fields?





Newtonian Physics, Calculus, differential equations, linear algebra, statistics and probabilities, java, c++, digital electronics, microprocessors, transistors, operational amplifiers, oscilloscope, proto board, and signal analysis(which includes time response of linear systems, Fourier transforms(Continuous, DTFT, FFT) and Laplace transform and sampling)


Thanks for your time and responses!!!!!!

 

[hevans1944]

You need to learn about video frame-grabbers and video image processing on a frame-to-frame basis. Basically, for each frame of video captured, you will identify pixels, or contiguous groups of pixels, that match the identification "dots" attached to the moving subject. The x-y locations of each of these pixels, or groups of pixels, is then stored in a memory array for processing in the next frame of video. You do this for each frame of video captured, so the array gets very large very quickly.

Between two successive frames of video you will need to develop an equation that relates the positions of the dot-tagged objects in the preceding frame to the same dot-tagged objects in the current frame, a tracking algorithm that predicts the position of each dot-tag in each frame of video. The problem is keeping track of the dot-tagged objects from frame-to-frame. They may overlap or be obscured by other parts of the moving object, so some creativity on the part of the software to "fill in" missing co-ordinates of the dot-tags will be necessary. When you are done processing as many frames as you have captured, the end result will be an array of trajectories for each dot-tag. You can try fitting the data to equations to allow interpolation between frames. This allows the frame rate to be slower, easing the load on the pixel processor.

This is not a trivial problem and it requires considerable processing power to accomplish it. I am not aware of any hardware/software that can accomplish the task in real time (yet!), but the PIXAR Image Computer (now defunct) came close and had the processing power to handle the job as a batch application. You should Google that and research its history and learn why it failed to gain market share. It was truly a machine ahead of its time in the 1980s.

A lot of the work in this area was initially supported by the I&R (intelligence and reconnaissance) community to aid photographic image analysts. It is desirable to be able to track changes in overhead imagery that occur over periods of days or even hours. Given the huge amount of data taken in the form of high-resolution imagery, and the limited number of human beings trained to analyze this data, it was inevitable that the government would look for an automated computer solution. This may have already occurred. The NSA and other agencies are rumored to have underground rooms filled with thousands of super-computers to handle the work. On the commercial side, Hollywood special-effects (SFX) post-production companies are invested heavily in specialized work stations and the highly trained and experienced users who perform the tasks of motion tracking, 3D modeling, ray tracing, and rendering that makes modern "action" motion pictures so exciting and realistic today.

If you want to learn more about this, become an IEEE member and join their SIGGRAPH special interest group. If you can afford it, plan to attend the 2015 SIGGRAPH full conference (not just the exhibits floor display!) in Los Angelas this year on August 9th to the 13th. Consider it an investment in your future.