Object Detection - Haar Cascade / LBP Cascade / HOG Descriptor - Result Value

For object deteciton results in:

  • Haar Cascade (Especially)
  • LBP Cascade
  • HOG Descriptor
    • anything else

The resullt of the detection is a number. For example for Haar Cascade, it is a floating point number (with fractinal part). e.g. 12.987654321. And also i have seen it range from at least some negative value (e.g. -1.0123) to a positive (e.g 4.519985).

So to evaluate how ‘good’ a result this is, what is the Minimum and Maximum possible values that the result can be (the range of the result)… Maybe It can be treated as a confidence value ?.

How is this value’s meaning to be interpreted when comparing the values of a detection produced from 2 cascade classifiers that each were trained to detect the same object (maybe they used different parameters, AND./OR training data, hence more than 1 classifier).

Also. how is this value’s meaning to be interpreted when comparing the values between 2 different cascade classifiers, each detecting differemt objects ?. (Can it be deduced that 1 classifier detection thier particular object more ‘surely’ than another, by comparing the value between the 2 calssifiers ?)

Also is there any meaning to the Negativeness of the value ? (Below 0)

please explain, how you get this number

For Haar Cascade (I remember), it is the levelWeights value from a version of the .DetectMultiScale() function:

http://docs.opencv.org/5.x/d1/de5/classcv_1_1CascadeClassifier.html#a1a5884c8cc749422f9eb77c2471958bc

1 Like

please explain, how you get this number

For Haar Cascade (I remember), it is the levelWeights value from a version of the .DetectMultiScale() function:

http://docs.opencv.org/5.x/d1/de5/classcv_1_1CascadeClassifier.html#a1a5884c8cc749422f9eb77c2471958bc