Statistical Multi-Objects Segmentation for Indoor/Outdoor Scene Detection and Classification via Depth Images
Adnan Ahmed Rafique, Ahmad Jalal, Kibum Kim
Abstract
With the advancement of technology, intelligence capabilities of machines are growing day by day. Researchers are committed to equip the machines with the capability of thinking humanly. Currently, the machines can sense and process information gathered from sensors. However, still there is a huge gape to improve the capability of thinking and understanding real scenes. Scene understanding is fiery area of research now a day. Therefore, we have proposed a model to understand and recognize a scene using depth data to make machines capable of interpreting the real time scenes like humans. The proposed recognition technique is a novel segmentation framework that uses statistical multi object segmentation to learn robust scene model and segregate the objects in the scene. Then, the unique features are extracted from these segregated objects to further process for recognition using linear SVM. Finally, multilayer perceptron is provided with the features and weights for the recognition of the scene. Our system demonstrated significant improvement over state-of-the-art systems. The proposed system is effective in autonomous vision systems like robotic vision, GPS based location finder, sports and security.