In the use of OpenCV hog + SVM for pedestrian detection reference when the big ideas and programs: https://blog.csdn.net/u011263315/article/details/41447545, is very much, but will be treated as positive and negative samples is good, the file path is still have after modification for their error as shown in the figure below:

Basically the error location is shown here:

For a long time, I did not know why it was wrong, and I consulted a lot of information but failed to solve it, which troubled me for a whole week. Finally, I solved the problem with the help of another big man, as follows:
At first, if we trace the alpha parameter, we find that it is not identified. Basically, there is no value assigned at all. At first, we thought that the problem was integrated from the ml.hpp file, and neither the existing.xml file nor the self-trained one had the value. Then the following a program about generating classifier and test files are commented out, to step through the part of sample training, found from the file name where you are to become very strange, basically to imread position could not read from the file image, then we do not have the right to read from the file to wrong image, according to this train of thought to make themselves the way in which to read file, specific procedures reference of the below: https://www.cnblogs.com/zhuwei1988/p/5746637.html, after the application is running through the test can also be trained classifier.
My environment is VS2013+OpenCV2.4.9, if other users have other errors, it is probably a version problem…
The programmer patiently read through the error line by line until he found the reason for the error… Record the process in order to avoid later have similar mistakes do not know how to debug……