Fig. 15From: Local shape feature fusion for improved matching, pose estimation and 3D object recognitionQualitative recognition results using single vs. ternary features. Detected objects are overlaid with different colors. Top row UWA scene 3, where the single features ECSAD, RoPS, and SI fail individually, but their ternary combination succeeds. Middle row Queen’s scene 22, where most of the single features show suboptimal results [false negatives in (e)–(h) and a false positive in (f)]. Bottom row all ten ternary combinations except NDHist+SHOT+SI (only four shown) detect all objectsBack to article page