Embedded System Design for Visual Scene Classification

Computer vision and robotics community is experiencing growing interest in visual scene classification due to availability of low cost and compact visual sensing devices. This paper presents framework aimed at embedded system design for visual scene classification. In the proposed framework we used data fusion of local and global descriptors as feature vectors for scene classification. We construct feature vector by integrating Local Quinary Patterns (LQP), Bag of Visual Words (BoW) and Histogram of Oriented Gradients (HOG). For classification multiclass Support Vector Machines (SVM) is used. Experiments are performed on publicly available MIT indoor scene classification database. Comparison of our approach with other methods show that our approach is efficient in terms of overall accuracy.