By Henk Heijnen, Project Manager at Technicolor Connected Home
Presented at SCTE’s Cable-Tec Expo 2019 on Monday, Sep. 30, 2019 in New Orleans
Listen to “SMART HOME & SMART NOTIFICATIONS: Multimodal event detection and notification”
As you read this, the latest wave of technologies is already being deployed to the connected home environment in markets across the globe. This paper details Technicolor’s research into smart notification strategies that leverage new and existing technologies in the home to capture and analyze environmental models. It provides insight into how network service providers (NSPs) can leverage the unique position of their customer premises equipment (CPE) presence to capture and process data about safety and security in the connected home environment and notify subscribers about anomalous events.
Solutions based on this research may provide service providers with an opportunity to help consumers manage conditions in the home by leveraging artificial intelligence (AI) and advanced audio capture technology. This paper details how in-home infrastructure can be harnessed by NSPs to develop sophisticated home monitoring services without compromising the security and privacy of consumers by:
- Leveraging Multimodal Sensors — By using multiple sensors such as temperature, humidity, pressure sensors for example devices to create models of both normal and abnormal situation and by integrating these various inputs with sound recognition (for example recognizing and therefore connecting fire/smoke alarms, but also glass breaking, dog barking, gun shots etc …), NSPs can capture and recognize unexpected or exceptional events — such as a fire or a break in — that then trigger user notifications. This strategy can replace the need for consumers to invest in expensive and complex dedicated security and safety systems.
- Delivering Highly Accurate Recognition Levels — by establishing a baseline context of a ‘normal’ situation so that the system can then capture anomalies that may represent an extraordinary event. This leverages the ability to receive and process information from multiple sensors and even hear disconnected devices — such as sound emitting alarms. It explores how artificial intelligence is being harnessed to develop a complex neural network model to achieve highly accurate recognition levels of 95-99% (audio recognition). This can greatly reduce false alarms.
- Protecting Subscriber Privacy — The system described in this white paper processes all information collected by sensors locally. Unlike many solutions that require constant remote monitoring from cloud resources, user information in this scenario never leaves the consumer’s premises — unless notifying a subscriber of a potential problem.