Deep Learning Method Based Binary Descriptor for Object Detection

摘要

The wide applications of the object detection techniques in the domains like video surveillance, security, military, automated industry tasks, biometrics has attracted the interest of the researchers. Deep learning is one of the most effective and efficient techniques for the object detection nowadays and has brought quite a revolution in this field. This paper proposes CNN architecture for the extraction of compact binary descriptors using stacked convolutional auto encoders without labeled data. PASCAL and CALTECH standard object datasets are used to validate the experimental results. The results are presented in terms of recall and precision performance matrices. The results show that the proposed architecture using CNN outperforms the rest of the state-of the art descriptor of its class. The recall and precision for the CALTECH dataset for the proposed CNN architecture is 0.98 and 0.93 respectively.