We revisit the problem of grey level corner detection to classify the techniques into two categories, namely template based techniques and intensity gradient based techniques. We generalize the measures for “cornerity” in each of these techniques and discuss the factors that affect the robustness of corner detection. We regard robustness as an important attribute in applications such as 3D depth estimation where the corners are used as features to be tracked over a sequence of images. We characterize three aspects of robustness, namely, detection, localization, and stability and discuss the performance of various cornerity measures with respect to these three criteria under our generalized framework. We describe a fusion based approach to incorporate robustness and implement a scheme based on this approach on the real-time PIPE (pipelined image processing engine) system.