Scale, Rotation and Pose Invariant 3D Object Detection using Log-polar Transform and Singular Value Decomposition

This paper proposed a novel biologically motivated 3D object detection algorithm which works robustly in the presence of large scale changes, rotation, occlusion, and pose changes. The robustness to scale and rotation is achieved through log-polar transformation procedure. Singular value decomposition (SVD) in conjunction with the distance-to-feature space measure is used to handle the pose variations and reduce the computational complexity. The system was evaluated using 3D object with varying scale and pose on different complex backgrounds. The experimental results show that the object detector was quite robust to all these variations