I’m getting more comfortable with PInvoke from C#. I’ve been using a web site that contains a pretty wide variety of recipes for getting at Win32 API calls with PInvoke.
At some point soon I need to take a look at the WindowsAPICodePack-Core which appears to have pre-built versions of some of these things. For now I’m happy that I’m getting closer to the point where I know how to invoke most API calls directly using PInvoke.
I do wish there was a more comprehensive reference document discussing all of the capabilities and ins and outs of using this facility. As is there are examples and specific documentation for some items (I’ve been using my copy of .NET 2.0 Interoperability Recipes: A Problem-Solution Approach to work out the basics and the PInvoke web site to extend that to more complicated examples.
I’ve put some of the sample code I’ve been playing with on GitHub at DupScan. This project is again code aimed at deduplicating file trees for archiving and management. The big driver here is the unique file ID API.
VHDL keeps coming up in places and my VHDL is more than a little rusty so I was back doing some refresher last night and will likely do some more tonight.
I need to get back to a point where I can read VHDL and make reasonable sense of it (and perhaps make small changes without breaking too much). If I hit the point where I’m feeling comfortable with it again I may dig out the Spartan-6 board I have lying around and see about trying some real work programming it.
This is something that keeps coming up, but once the need fades off I find other things that are higher priority and never get past the early stages…need to reach basic fluency this time around.
I built a few more beacons to play with last night. I’ve got some CR2023 battery holders that are smaller than the AA cell battery clips and have integrate power switches that I’m using for these. The red and blue LEDs work as expected (though they look dimmer to my eye than the white one on the first version we built).
I put together an IR LED based beacon as well (though without the diffuser as I’m not sure the diffuser plastic is transparent at 970 nm). I’ve demonstrated that the cameras I’ve got can see the IR LEDs on a remote control, even with their IR filters intact. I could see that the IR LED was on, but its brightness was much less than the brightness of the visible light LEDs so no real advantage to going with IR.
I also finished updating my RPi machines to the ‘scratch’ OS image and am close to having all of them built for OpenCV libraries. Once I get the last machine loaded up, I expect to switch back to one of the Ubuntu NUC machines and write some code to read from cameras and process the result.
Attaching LED information for easy later location…
Played with some commercial motion capture software last Friday, more here.
I’ve upgraded most of my RPi machines to scratch and I believe I have the process of building OpenCV 4.0.1 on them to the point where it is reliable.
I have realized that I can’t push the full, unprocessed output of these webcams over the ethernet links. I’ll have to pre-process the data to reduce total size in order to make things work. Not sure whether the ARM CPUs have the necessary performance or not. I’ll have to look at this and see what I can see.
This time I pasted all of the pieces into a single shell script (instead of running it piecemeal) and things went smoothly. Still took a while, even on the Core-i5 system with an SSD and 32 GB of physical memory. I specifically pulled 4.0.1 from git rather than the default choice from the source of the instructions that built ‘master’.
Tonight I’ll have to write a bit of code to use the library and see if I can get streams of images programmatically from one or more of the webcams I’m working with.
At some point I may bump up the swap space on another of my RPi systems to see if the same script works there as well.
Well…almost got OpenCV building on my main Ubuntu machine at home. I was copying fragments from the directions on the OpenCV site into an SSH session and clearly missed something along the way. I’ll have to build a single, large shell script up front next time and then run that. Unfortunately it seems as if (I may learn better later on) once CMake has done its magic, lots of things get baked into the files that drive the build in ways that really, really want a rebuild if things don’t go quite right.
I was working from the instructions here. (Other instructions here).
I was surprised to see that the github repo and contrib did not have a branch tag for release 4. It looks as if there’s a stable release out there, but 3.4 and master seemed like the available choices.
The main site clearly indicates that version 4 has been released with pre-built windows and ios downloads and documentation. I’m not sure currently how to pull that stable release code from git though.
I expect to take another run at the 4.0 build on Ubuntu tonight. Other than script grabbing issues, the build went smoothly…32 GB of memory and a Core-i5 CPU work better than an low-end ARM and 1 GB swapping on a micro-SD card.
Hmm…more github presence for version 4 here and here. Looks like OpenCV 4.0.1 is the latest. Ah…tags not branches here…need to look at pulling the appropriate tag for the build. Easy enough…just list the tags, find the 4.0.1 tag and check that out.
The CMake run for the C driver seems to always pull in cygwin headers that don’t work. I had to manually remove them from the build configuration. I had to disable the ICU (also not present). I added the installer that is used. Once these were done, things almost completely built and installed.
I’ve had to select the obsolete python 2 interpreter as the default for the C++ drivers and hack the version for the C drivers to 0.0.0 (I’m guessing this is related to the python problems as that code seems to pull version numbers for the packages).
Building the C++ drivers now and hoping this goes well. I’ll be using the 64 bit drivers if all of this works. Being able to push to MongoDB from C and C++ code will be very helpful if this all finishes and runs.
Building C++ supporting libraries on my small dev machine. Looking at another pass of C++/CLI work and some more thinking about motion capture options.
I’ve built zlib and bzlib2 yesterday. To get these building with VS 2017 I built them with their standard build procedures (which appeared to build 32 bit libraries) and then created visual studio 2017 projects to build the same pieces using fresh settings.
I’ve got boost, openssl, opencv and the mongodb c and c++ libraries on my list for today. The mongodb libraries are my primary target here though all of the above are of interest. This is also what is driving my CMake reading as several of these libraries use CMake for their builds.
I’m pretty interested in C++/CLI as a way to get access to C and C++ functionality from C#. Given the productivity that C# provides, access to C and C++ APIs and libraries from that environment would help quite a bit. My primary driver was access to the Win32 file ID API for some file management work, but Win32 APIs keep coming up now and again. I do need to take a look as some newer C# libraries that Malcolm suggested that may provide pre-packaged access here.
In the past, I’ve run into build issues when combining C# and C++/CLI in the same project. I expect this was setup issues so I’m now looking to take a more serious look at this.
The motion capture work will start out with OpenCV and getting my web cams running with code behind them to process the images. I’m thinking that pre-recording things and then post processing might be a good idea…using a light strobe to synchronize things could simplify time sync as well. Much to consider…
I’ve got boost, zlib and libbz2 built on boojum now. Next step will be to test the installs with a small program or two. Hoping that all went well and I have usable, native, win64 libraries available.
Once those have been checked out I’ll take another run at building the MongoDB drivers…this time on boojum rather than chaos. Boojum has a shorter history and is distinctly less cluttered than chaos so I’m hoping that this just works. Expecting less than that though…I’ll probably need to dig deeper into he CMake configs to get where I want to go. I’ll update as I move forward…now off to lunch though.
We tried doing some motion capture with a kinect 2 today. Results were mixed though less than ideal.
I am thinking of using OpenCV to acquire images and use them to register locations in three-space. I currently have two high quality cameras and a lower quality one to play with…I expect that if things go well I’ll likely pick up a third higher quality camera to work with along with some hard-mounts for them in the basement. Ideally I’d love to be able to code up some decent motion capture functionality using cameras and reference marks on limbs.
A bit of experimentation will be required in order to get there though.
I am wondering whether a faster, lower resolution webcam might do better for this. There are some $20.00 a piece cameras with USB-2 interface out there (Microsoft Lifecam-3000 or Logitech C270) instead of the higher-end, USB-3 autofocus Logitech C930e. Both wondering whether the lower resolution and USB-2 may provide a higher frame rate and whether fixed focus will stay consistent without periodic refocus hits.
I started work as a Principal Software Engineer at KMC systems in April of 2016. I'm enjoying the work and the team there so far.