Towards Real Time Hardware Based Human and Object Detection: A Review

Abstract

Effective solution to image based generic human and object detection problem can potentially benefit many usecases such as autonomous vehicles (self-driving cars) and surveillance applications at scale. While considerable advancement is made in machine/deep learning based solutions to these problems, their training and testing requires enormous amount of compute, hindering their widespread adoption in power constrained environments such as cars, surveillance cameras, drones or remote sensing systems etc.In this paper, some of the deep learning computer vision solutions namely Convolutional Neural Network (CNN) is reviewed for human detection and object detection in general. In addition, we provide detail of datasets used and hardware architecture used for training and evaluation of these algorithms. In particular, the usage of software (GPUs) and hardware such as Field Programmable Gate Arrays (FPGAs) as a potential alternative to traditional hardware to deploy these deep/machine learning systems in real time has been reviewed.