HOG特征相似度测量？

Question

I am trying to implement particle filter for detection-based-tracking, currently trying to update the weight based on appearance model. Before tracking, I have a detector based on HOG+SVM, which means currently, I have HOG vector for each detected person. For the new generated particle, I want to set the weight based on the similarity of HOG vector compared with the detector's HOG vector. So any suggestion for algorithm which could measure the similarity of HOG vectors? Thanks

Answer 1

Try L2 metric (ordinal distance between feature vectors) or cosine distance.

double CosineDistance(float* v1, float* v2, size_t count)
{
    double dot = 0.0, denom_a = 0.0, denom_b = 0.0;
    for (unsigned int i = 0u; i < count; ++i) 
    {
        dot += v1[i] * v2[i];
        denom_a += v1[i] * v1[i];
        denom_b += v2[i] * v2[i];
    }
    return dot / (sqrt(denom_a) * sqrt(denom_b));
}