Embedded Computer Vision for Privacy-Preserving Real-Time Analytics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computer vision systems for home environments are either too simple and unpredictable or too complex and uneconomical, with high computational and storage costs that scale linearly with the number of users, cameras, and resolution, lacking accuracy and scalability, and failing to provide real-time processing and privacy guarantees.
Innovation Solution
A computer-vision system that generates a digital representation of people or objects from pixel streams, determines attributes, and controls networked devices, using an ASIC-based engine for real-time metadata processing without continuous video output, enabling reliable tracking and control of multiple objects with varying orientations and gestures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sophisticated video analysis is performed on servers, then analysis accuracy is improved, but scalability and cost are worsened due to linear scaling of storage and computational costs
Solution Approach 1:
The system segments the video processing workflow into two distinct parts: (1) simple real-time processing performed locally at the camera using dedicated vision processors, and (2) sophisticated analysis performed selectively on extracted key frames or events at the server. This segmentation allows the system to achieve high analysis accuracy for critical moments while avoiding the need to transmit and process entire video streams, thus breaking the linear scaling relationship between user/camera count and computational cost.
Solution Approach 2:
The system extracts only the most relevant information from video streams - specifically key frames containing detected events, anomalies, or important moments - and transmits only these extracted elements to the server for sophisticated analysis. This extraction approach eliminates the need to process complete high-resolution video streams on servers, dramatically reducing storage and computational requirements while maintaining analysis accuracy for critical events.
2Productivity
If simple video analysis is performed in cameras, then processing speed is improved, but analysis reliability and value are worsened
Solution Approach 1:
The system introduces a dedicated vision processor as an intermediary component between the camera sensor and the server. This specialized hardware performs real-time pre-processing, event detection, and key frame extraction with high reliability and consistency. The vision processor acts as a mediator that prepares standardized, reliable data for server analysis, ensuring that the simple real-time processing at the camera level produces predictable and valuable results that can be reliably acted upon.
3Measurement precision
If full-frame video is transmitted to remote servers, then analysis capability is improved, but bandwidth consumption and cost are worsened
Solution Approach 1:
The system extracts only essential video information - key frames representing detected events, anomalies, or important moments - and transmits only these extracted elements to the server. This extraction is performed by dedicated vision processors that identify and capture only the most relevant moments in the video stream, reducing data transmission volume from complete high-resolution video streams to selective key frames while preserving full analytics capability for critical events.
4Measurement precision
If high-resolution video processing is performed, then detection accuracy is improved, but computational cost and power consumption are worsened
Solution Approach 1:
The system segments the computational workload by resolution and processing requirements: dedicated vision processors handle high-resolution real-time processing for object detection and event identification, while servers perform lower-resolution sophisticated analysis on selectively extracted key frames. This segmentation allows high detection accuracy to be achieved at the camera level using specialized hardware, while reducing the computational burden and power consumption at the server level by processing only essential extracted moments rather than complete high-resolution streams.
Data Source
AI summary
The field of the invention relates to computer vision systems and methods providing real time data analytics on detected people or objects in the home environment or other environments. It is based on an embedded engine that analyses an image from a raw sensor and virtualised the image into a digital representation enabling a digital understanding of the environment while guarantying privacy. It comprises multiple image processing blocks and embedded firmware.