Glimpse: Continuous, Real-Time Object Recognition on Mobile Devices

• Main Authors: Chen, Tiffany Yu-Han (Author), Ravindranath, Lenin (Author), Deng, Shuo (Contributor), Bahl, Paramvir (Author), Balakrishnan, Hari (Contributor)
• Other Authors: Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science (Contributor)
• Format: Article
• Language: English
• Published: Association for Computing Machinery, 2017-07-18T15:33:37Z.
• Subjects: Article
• Online Access: Get fulltext

 Glimpse is a continuous, real-time object recognition system for camera-equipped mobile devices. Glimpse captures full-motion video, locates objects of interest, recognizes and labels them, and tracks them from frame to frame for the user. Because the algorithms for object recognition entail significant computation, Glimpse runs them on server machines. When the latency between the server and mobile device is higher than a frame-time, this approach lowers object recognition accuracy. To regain accuracy, Glimpse uses an active cache of video frames on the mobile device. A subset of the frames in the active cache are used to track objects on the mobile, using (stale) hints about objects that arrive from the server from time to time. To reduce network bandwidth usage, Glimpse computes trigger frames to send to the server for recognizing and labeling. Experiments with Android smartphones and Google Glass over Verizon, AT&T, and a campus Wi-Fi network show that with hardware face detection support (available on many mobile devices), Glimpse achieves precision between 96.4% to 99.8% for continuous face recognition, which improves over a scheme performing hardware face detection and server-side recognition without Glimpse's techniques by between 1.8-2.5×. The improvement in precision for face recognition without hardware detection is between 1.6-5.5×. For road sign recognition, which does not have a hardware detector, Glimpse achieves precision between 75% and 80%; without Glimpse, continuous detection is non-functional (0.2%-1.9% precision).