LAUSR

Recognizing indoor scene and objects and estimating their poses present a wide range of applications in robotic field. This task becomes more challenging especially in cluttered environments like the indoor scenery. Scaling up convnets presents a key component in achieving better accuracy results of deep convolutional neural networks. In this paper, we make use of the rethinked efficient neural networks and we fine-tune them in order to develop a new application used for indoor object and scene recognition system. This new application will be especially dedicated for blind and visually impaired persons to explore new indoor environments and to fully integrate in daily life. The proposed indoor object and scene recognition system achieves new state-of-the-art results in MIT 67 indoor dataset and in scene 15 dataset. We obtained 95.60% and 97% respectively as a recognition rate.