Human detection in surveillance videos using MobileNet

Abstract

Video surveillance is of paramount importance. Surveillance systems are being developed to perform surveillance tasks automatically. Human detection process allows to build effective surveillance system and several approaches exist in literature for detection tasks that can be divided mainly in traditional machine learning approaches. The learned features are extracted automatically. They give most accurate results in image recognition tasks but they need more computing power and large space memory which is challenging for embedded devices. Ex: VggNet, ResNet. In this paper, we used MobileNet deep convolution neural network with transfer learning approach to build deep learning model for human classification. We used INRIA person dataset to train and test our model. We achieved a good accuracy and comparative precision.