A Computational Offloading Framework for Object Detection in Mobile Devices

Recently, the role of mobile devices has changed from a calling or entertaining device to a tool for making life easier. However, this growing role is associated with extensive computing requirements to execute tasks such as object detection. Moreover, executing heavy tasks in a mobile with limited resources takes a long processing time and consumes much energy. This paper presents a computational offloading framework to improve the performance of object detection tasks. The framework uses a simple decision model to select between local processing and offloading based on the network context. A demo has been developed to evaluate the framework performance. The experimental work includes different image sizes and employs 3G and Wi-Fi networks. The results show a response time speedup that could reach five times for small images and 1.5 times for big images. The energy saving also ranges from 80% to 50%. Furthermore, offloading through a Wi-Fi network shows more performance stability than a 3G network. Finally, results demonstrate that offloading the object detection computation decreases the memory allocations to less than 1% in comparison with local processing.