The work aims to improve the detection efficiency of battery production enterprises and avoid false detection to reduce production cost of the enterprise. The deficiencies in traditional sorting methods were analyzed. An industrial robot sorting system based on machine vision was constructed with ABB parallel robot and American Connal intelligent camera as platform. The classic SIFT algorithm was used to process the image information of sorted objects, which overcame the adverse impacts of illumination and position change on visual system and improved the sorting efficiency. ABB parallel robot could identify objects to be sorted according to image information. It worked stably and reliably and enjoyed success rate of 100%. The software and hardware of the sorting system are reasonably designed and meet the requirements of battery manufacturers.