Let's write a master's thesis on Kalman target tracking on cnki. There are many, of course, journals can also be used, but generally speaking, master's thesis can be more detailed, and then study one carefully. This is basically no problem in theory. Programming is done with MATLAB, and it is very troublesome to use C. Many algorithms are not compiled from scratch. Matlab integrates many algorithms, just find out and call them.

The idea of Kalman tracking is this:

0. If your video is actually recorded, in order to prevent false targets from being detected, the input image should be filtered first, and there is a simple median filtering.

1. The motion region can be obtained by using the background difference or the inter-frame difference for the video sequence. The emphasis here is background modeling. If it's too much trouble, don't look at what is single Gaussian or multi-Gaussian. As long as an empty frame (without moving target) is found as the background, the moving region will be obtained after the difference, and then binarization is convenient for later processing. Then, according to whether there is shadow or not, the shadow removal work is carried out.

2. The above step is to detect the motion area. According to your detection, it is necessary to draw a circle around the edge. In this paper, how to extract the edge points of the target on matlab is studied. Just change the edge points to the same pixel values on the in-situ Tu Tu, and the detection is over.

3. Tracking, we must first find the center of the target, because the target is not just a pixel, it must have a center to represent its coordinate position. This method is our own idea. The center of the circumscribed rectangle can be found by averaging, and then each frame will have a series of center coordinates.

4. Kalman, the role of Kalman is still mainly filtering, which is equivalent to treating all the coordinates in the third step as a signal sequence, filtering with Kalman, and filtering while detecting. Kalman mainly remembers those five formulas, and can basically make them up by knowing their recursive process. As for the filter parameters, please look them up in Resources. There are not many lines in the program compiled by Kalman. Don't be afraid.

5. If it is multi-target tracking, target matching is equivalent to detecting two targets per frame. You need to know which target in the previous frame corresponds to each target in the latest frame.

The above gives you a general idea. What kind of job do you choose according to your own tasks? This topic is not difficult, so be confident.