Automatic Indoor-Outdoor Scene Recognition With Supervised Machine Learn¬ing Algorithm

Abstract

In computer vision, Indoor-outdoor scene recognition plays an important role to identify the type of scenes for service robots, so as to interact with real-time environments. The proposed system utilizes feature extraction and classification techniques as two stages to recognize the scene type. In the first stage, SIFT, SURF, and ORB algorithms are used to extract the features based on the N number of key points from indoor and outdoor images. In the second stage, those extracted 128*N numbers of features from SIFT, 64*N numbers of features from SURF, 32*N numbers of features from ORB are given as an input to supervised machine learning algorithms like Random Forest and SVM. The performance of the proposed system is evaluated using MIT indoor67 and scene15 dataset and the proposed system outperforms well than the other existing systems. From the evaluated results, SURF+Random Forest outperforms well and achieves the recognition rate of 98% with an error rate (ER) of about 0.2%, SURF+SVM achieves the recognition rate of 97.23% with an error rate (ER) of about 1.77%.