AME Meteor Studio Research on Energy-Efficient Computer Vision by Jinhan Hu and Venkatesh Kodukula
Advances in vision processing have ignited a proliferation of mobile vision applications, including augmented reality. However, limited by the inability to rapidly reconfigure sensor operation for performance-efficiency tradeoffs or exploit near-sensor processing, high power consumption causes vision applications to drain the device’s battery. To explore the sensor resolution reconfiguration latency bottlenecks, we profile it on mobile devices and find three major sources of latency in current operating systems. Based on our intuitions, we propose a redesign of Android camera system which enables smooth transitions between sensor configurations. We also investigate the thermal bottlenecks to processing vision tasks near the sensor and propose investigations to unlock the potential of near-sensor processing.
Jinhan Hu is currently a doctoral student in the School of Electrical, Computer and Energy Engineering at Arizona State University. Jinhan completed his Master of Science at Auburn University and his bachelor's degree at Chongqing University. His work concentrates on optimizing operating systems for sensing and actuation on mobile devices.
Venkatesh Kodukula is currently an Master of Science student in computer engineering at Arizona State University. Venkatesh completed his bachelor's degree at the National Institute of Technology, Warangal. His current work concentrates on the thermal dependencies of near-sensor vision processing.