This paper presents a new parallel binocular vision algorithm implemented on an associative memory. The proposed method is accurate against segmentation errors because it uses Hough-Transform-based segmentation. In addition, we introduce a new Generalized Hough Transform algorithm for segment matching using simultaneously the point segments and their local gray intensity. This has the merit of eliminating the ambiguity obtained by methods using segment similarity. Experimental results demonstrate the robustness of our algorithm against noise. Moreover, the complexity evaluation for an image sized 256×256, independently of the number of edge points, indicates that a single PC board containing a pair of two Content Addressable Memory (CAM) chip LSIs is sufficient for supporting this application in real-time. This is an improvement of previous systems, where execution time of the second-order using a greater amount of hardware has been proposed.