Investigating the Importance of Shape Features, Color Constancy, Color\n Spaces and Similarity Measures in Open-Ended 3D Object Recognition

Abstract

Despite the recent success of state-of-the-art 3D object recognition\napproaches, service robots are frequently failed to recognize many objects in\nreal human-centric environments. For these robots, object recognition is a\nchallenging task due to the high demand for accurate and real-time response\nunder changing and unpredictable environmental conditions. Most of the recent\napproaches use either the shape information only and ignore the role of color\ninformation or vice versa. Furthermore, they mainly utilize the $L_n$ Minkowski\nfamily functions to measure the similarity of two object views, while there are\nvarious distance measures that are applicable to compare two object views. In\nthis paper, we explore the importance of shape information, color constancy,\ncolor spaces, and various similarity measures in open-ended 3D object\nrecognition. Towards this goal, we extensively evaluate the performance of\nobject recognition approaches in three different configurations, including\n\\textit{color-only}, \\textit{shape-only}, and \\textit{ combinations of color\nand shape}, in both offline and online settings. Experimental results\nconcerning scalability, memory usage, and object recognition performance show\nthat all of the \\textit{combinations of color and shape} yields significant\nimprovements over the \\textit{shape-only} and \\textit{color-only} approaches.\nThe underlying reason is that color information is an important feature to\ndistinguish objects that have very similar geometric properties with different\ncolors and vice versa. Moreover, by combining color and shape information, we\ndemonstrate that the robot can learn new object categories from very few\ntraining examples in a real-world setting.\n