To http://blog.csdn.net/pozen/article/details/7023742
Discriminatively trained deformable part models URL of felzenszwalb: http://www.cs.brown.edu /~ Pff/latent/
It is said to be the best object detection method at present. I tried it myself, and the effect was really good. However, the Code can only run on UNIX, Linux, or Mac.
(The methods used by Pascal VOC to win the OBJ detection championship in the last two years are based on this framework, but others' research is not open to the public. By the way, this is not true)
However, you only need to make some modifications to run on Windows:
1. Add the following statement to DT. CC: # define int32_t int
2. Add the following content to features. CC & resize. CC:
# Define bzero (a, B) memset (A, 0, B)
Int round (float a) {float TMP = A-(INT) A; If (TMP >= 0.5) Return (INT) A + 1; else return (INT) ;}
3. In resize. CC: alphainfo OFS [Len]; this sentence is changed to: alphainfo * OFS = new alphainfo [Len]. Of course, delete [] OFS is added after the same scope.
4. In compile. M: The end is added with Mex-O fconv. CC.
% Use one of the following depending on your setup
% 1 is fastest, 3 is slowest
% 1) multithreaded convolution using Blas
% Mex-O fconvblas. CC-lmwblas-O fconv
% 2) mulththreaded convolution without Blas
% Mex-O fconvmt. CC-O fconv
% 3) Basic convolution, very compatible
% Mex-O fconv. CC-O fconv
Mex-O fconv. CC
Several other fconvs cannot run on windows with multithreads from other platforms!
After changing the above several points, you can run it. Run demo. m to check the effect ,,,