A DNN-based object detection system on mobile cloud computing
Authors:
- 1 September 2017
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
Abstract
With the development of big data and the improvement of computing power, deep learning has made a very prominent breakthrough in computer vision. However, the computational overhead of the Deep Neural Network (DNN) for video processing in mobile devices is extremely high. To address the problem above, this paper combines smartphones with the cloud to realize a DNN-based object detection system. The main contributions are three-fold. (i) A model scheduling algorithm is proposed to adaptively select the operating environment (cloud or mobile) according to the conditions of network and mobile devices. (ii) To meet the hardware requirements of mobile devices, the compact model variants are trained and generated with a small loss of precision. (iii) To reduce the latency, the outputs of DNN models are used to process (add bounding boxes and annotations) the video directly. Test results for runtime and precision show that our system outperforms the state-of-the-art in both detection accuracy and running speed.Keywords
This publication has 7 references indexed in Scilit:
- Resource-constrained implementation and optimization of a deep neural network for vehicle classification2016
- CNNdroid2016
- MCDNN2016
- You Only Look Once: Unified, Real-Time Object Detection2016
- DeepX: A Software Accelerator for Low-Power Deep Learning Inference on Mobile Devices2016
- Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation2014
- Vision meets robotics: The KITTI datasetThe International Journal of Robotics Research, 2013