Rotation angle recovery for rotation invariant detector in lying pose human body detection

A method for rotation-invariant lying-pose human body detection in overlooking images is proposed. The rotation-invariant histogram of oriented gradient using Fourier analysis in polar coordinate is exploited as descriptor for lying-pose human body. And then the authors used the exhaustive sliding window search strategy with multiple scale scan to localise human body. Finally, principal component analysis (PCA) is used to determine the rotation angle of the exhaustive sliding window based on the classifier output scores. Experiments on their built XiaMen University Lying-Pose Dataset (XMULP) show the effectiveness of their proposed method.