Software Framework for testing binocular vision

The test platform of binocular vision includes two parts: software and hardware.

Here we only consider this part of the software.

For binocular vision, camera calibration is provided, matching points are found, and finally the binocular vision algorithm is used. To verify the binocular vision algorithm, of course, we need to build a software platform, that is, capturing video streams in real time, and then using the binocular vision algorithm for real-time processing, obtain specific results (generally used for ranging, such as the distance of a moving object ).

Because the algorithm in binocular vision may be transplanted to another platform in the future, the design of this framework for binocular vision test program should also take this into account.

This is a dialog box-based program. The entire program is roughly divided into three parts. One part is used for capturing video streams from Two USB cameras. Part is the algorithm part of binocular vision. Another part is Synchronization Control for multithreading. In addition to the automatically generated class in the dialog box, there are also three classes in this program. Two classes are used to intercept the Video Stream Image of the camera (one is the creation of filter graph, the other is the callback setting at the time), and the other is used to implement the binocular vision algorithm.

During the compilation of this program, DirectShow is used for camera image capturing. I can write this program within a week. I don't think I could have imagined it before. Maybe it's a brave man under pressure, and maybe it's also something I 've accumulated. The difficulty of writing this program framework lies in the two cameras capturing images at the same time for real-time processing, as well as the future convenience of porting and algorithm writing, there are a lot of things to consider in the design of the program framework. In the framework of the program, I think, I am not doing very well, although, now we can easily compile and transplant algorithms. I hope you will have a good look at the books on Design Patterns in the future.

When using DirectShow to write a camera to capture video streams and capture images, I used many examples of Microsoft program. Now there is another small problem with this program, that is, it still cannot get the Frame Rate of image processing. At this point, after this week, I will take a good look at it and it should be able to solve it. This week should mainly focus on image processing algorithms. Otherwise, the task may not be completed on time.

In addition, because image processing and computer vision algorithms are not well done yet, testing the framework of this program is still needed, I just asked to save the BMP image instead of the binocular vision algorithm, and the result experiment was successful (I can save a lot of BMP images continuously, of course, in my need)