Indoor Scene and Position Recognition Based on Visual Landmarks Obtained from Visual Saliency without Human Effect

Abstract

Numerous autonomous robots are used not only for factory automation as labor saving devices, but also for interaction and communication with humans in our daily life. Although superior compatibility for semantic recognition of generic objects provides wide applications in a practical use, it is still a challenging task to create an extraction method that includes robustness and stability against environmental changes. This paper proposes a novel method of scene and position recognition based on visual landmarks (VLs) used for an autonomous mobile robot in an environment living with humans. The proposed method provides a mask image of human regions using histograms of oriented gradients (HOG). The VL features are described with accelerated KAZE (AKAZE) after extracting conspicuous regions obtained using saliency maps (SMs). The experimentally obtained results using leave-one-out cross validation (LOOCV) revealed that recognition accuracy of high-saliency feature points was higher than that of low-saliency feature points. We created our original benchmark datasets using a mobile robot. The recognition accuracy evaluated using LOOCV reveals 49.9% for our method, which is 3.2 percentage points higher than the accuracy of the comparison method without HOG detectors. The analysis of false recognition using a confusion matrix examines false recognition occurring in neighboring zones. This trend is reduced according to zone separations.