Edge AI and Vision Insights: April 20, 2021 Edition

LETTER FROM THE EDITOR

Dear Colleague,

Edge Impulse, Intel and Qualcomm: three companies that will be showing up in a big way at the Embedded Vision Summit on May 25-28. Each company will be sending its top engineers, tech experts and top executives to bring you key insights on the latest hot topics in computer vision and AI—and in some interesting ways. See a sample of sessions from these companies below. For the full program schedule and abstracts, visit the 2021 Summit Program and Schedule pages. And then register for the Summit today!

Edge Impulse

Intel

Qualcomm

Brian Dipert
Editor-In-Chief, Edge AI and Vision Alliance

VISION-BASED ACTIVITY ANALYSIS

Multi-modal Re-identification: IoT and Computer Vision for Residential Community Tracking
The recent COVID-19 outbreak necessitated monitoring in communities such as tracking of quarantined residents and tracking of close-contact interactions with sick individuals. High-density communities also have many non-resident visitors (delivery, repair, social) and tracking allows safeguarding of sensitive areas like playgrounds. Vision techniques for face recognition and person tracking are severely challenged in real residential communities; for example, poor accuracy for children and the elderly due to sparse training data and suboptimal positioning of cameras. Also, attributes such as whether somebody is sick, or is a visitor, cannot be observed by a camera. In this talk, Kit Thambiratnam, former General Manager of the Seedland AI Center, introduces how Seedland builds safer and healthier residential communities in China using cross-modal IOT-plus-vision person tracking and intelligence. He describes his company’s solution, which propagates intelligence from smart community devices and services across camera-based person-tracking to build a rich annotated graph of behavior and attributes. Thambiratnam details technical solutions to real-world challenges in person identification, annotation and multi-day tracking.

Video Activity Recognition with Limited Data for Smart Home Applications
Comcast’s Xfinity Home connects millions of home smart cameras and IoT devices to improve its customers’ safety and security. The company’s teams use computer vision and deep learning to understand video and sensor data from these devices to identify relevant events so that it can improve the user experience. Specifically, Comcast has explored the spatial-temporal relationships among objects, places and actions. The company has also developed a semi-supervised learning approach for video classification (VideoSSL) to detect certain activities using limited training data. Using these techniques, and as described in this presentation from Hongcheng Wang, Director of Technical Research at Comcast, it has achieved very promising results on activity recognition with multiple datasets.

DRIVER MONITORING AND EYE TRACKING

Driver Monitoring Systems: Present and Future
Monitoring drivers and passengers inside of vehicles is an increasingly critical capability. For example, driver monitoring is required in order for cars to obtain a top safety rating from NCAP. This presentation from Petronel Bigioi, CTO for Imaging at XPERI, introduces XPERI’s driver and in-cabin monitoring solutions and examines real-world use-cases in which these solutions are being deployed. Bigioi illustrates the evolution of these technologies as they have been used in conventional cars, as they are increasingly being used in cars with partial self-driving capability, and how they are likely to be used in fully automated vehicles. This goes well beyond driver monitoring to include new types of safety features as well as non-safety uses such as entertainment, personalization, human-machine interfaces and even monitoring occupant health.

Eye Tracking for the Future
Eye tracking is an increasingly important technology for applications ranging from augmented and virtual reality head-mounted displays to automotive driver monitoring. In this talk, Peter Milford, President of Parallel Rules, introduces eye tracking techniques and technical challenges. He also explores camera and computational requirements for eye tracking, and highlights selected use cases and applications.