4d Implicit Shape Model For Rotation Invariant Contour-Based Object Detection

In this paper, we present a rotation invariant contour-based object detection method. An Implicit Shape Model (ISM) is learnt utilizing local contour features. As opposed to the original ISM, in which voting is performed in a 3D location-scale voting space, we propose a 4D ISM which estimates the in-plane rotation angle of an object as the fourth dimension of the voting space in addition to its location and scale. Experimental results on a data set containing rotated objects demonstrate the capability of our proposed approach in localizing rotated objects.