OPTIMIZATION OF DEEP LEARNING FOR HUMAN DETECTION

Abstract

The main objective of the paper is human detection based on optimized convolutional neural networks (CNN). In future, this is very important for self-driving cars, surveillance systems, gender classification and other applications. In this paper, we propose an algorithm to detect and classify images as human or non-human. The algorithm is Faster R-CNN with ResNet50 Network model. The object localization done by using Faster R-CNN with ResNet50 and optimization of ResNet50 is used for classification of object.  Here,ResNet50 Convolutional Neural Networks is optimized by use of stochastic gradient descent with momentum (SGDM). Optimization of deep learning is implementing for increase the accuracy of human detection with use small amount of dataset and to reduce the training time. The proposed system will be fast enough to detect and recognize human.