traincascade based on HOG feature

Question

as we know， the opencv traincascade can handle all the three type features HAAR HOG and LBP I have already study the insight of HAAR and LBP features adapt AdaBoost, but I don't understand HOG part: 1. how the program handle the HOG features? 2. what's the weak classifier in HOG features? the hog operators or block, cell, bin informations? 3. any one can explain the .xml file for hog detector?<stages> <!-- stage 0 --> <_> <maxWeakCount>9</maxWeakCount> <stageThreshold>-1.8159391880035400e+000</stageThreshold> <weakClassifiers> <_> <internalNodes> 0 -1 206 1.9242146983742714e-002</internalNodes> <leafValues> -3.4242480993270874e-001 -8.9670753479003906e-001</leafValues></_> <_> <internalNodes> 0 -1 72 1.3402653858065605e-002</internalNodes> <leafValues> -7.8335583209991455e-001 4.7632049769163132e-002</leafValues></_> <_> <internalNodes> 0 -1 228 4.8146050423383713e-002</internalNodes> <leafValues> 1.7565745115280151e-001 -6.2845951318740845e-001</leafValues></_> <_> <internalNodes> 0 -1 97 1.4250885695219040e-002</internalNodes> <leafValues> -7.5739115476608276e-001 1.9031114876270294e-001</leafValues></_>