In this work, the machine vision technique is introduced into an automated beverage bottle reclaiming system. Based on the characteristics of beverage bottles made of PET materials, local features and geometrical invariability, we extract the corner points at subpixel accuracy, define the curve angle around the central corner and its sign as the eigenvalues of a beverage bottle, and propose a blockmatching algorithm. Experimental results show that the algorithm is stable, and has good discrimination capability, high recognition rate and fast computation speed. It can meet the requirements of reclaiming beverage bottles accurately in real time.