Removable Media Deep Learning Accelerator for Surveillance Camera Video Analytics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Surveillance cameras lack the capability for efficient video analytics and data compression, relying on traditional methods that consume high energy and computation time, and do not leverage advanced AI techniques effectively.
Innovation Solution
A removable media with a deep learning accelerator (DLA) and random access memory is introduced, enabling surveillance cameras to perform computations for artificial neural networks (ANNs) autonomously, analyzing video data to generate analytics and compressing video files using deep learning-based techniques, reducing energy consumption and computation time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional video processing methods are used in surveillance cameras, then device complexity is reduced, but productivity and energy efficiency deteriorate due to high computation time and energy consumption
Solution Approach 1:
The system segments video processing into two parts: traditional compression (H.264/H.265) handled by existing camera hardware, and AI-based analytics (object detection, recognition) handled by the removable media with DLA. This segmentation allows each component to optimize for its specific function, improving overall productivity while managing energy consumption efficiently.
Solution Approach 2:
The removable media acts as an intermediary between the surveillance camera and the cloud/server. It performs AI computations locally using the DLA, generating analytics that are then transmitted to remote systems. This intermediary approach reduces the need for continuous high-bandwidth cloud connectivity and enables real-time local decision-making.
2Measurement precision
If AI-based video analytics are implemented, then measurement precision and productivity improve, but device complexity increases due to additional processing requirements
Solution Approach 1:
The complex AI processing functionality is extracted from the main surveillance camera system and placed in a separate removable media component. This allows the camera to maintain its core surveillance functions while gaining advanced analytics capabilities through the removable module, which can be easily upgraded or replaced without affecting the main system.
Solution Approach 2:
The removable media with DLA is designed to be universally compatible with multiple camera models and can perform multiple AI functions (object detection, facial recognition, behavior analysis). This multi-functionality reduces the need for separate specialized devices and simplifies the overall system architecture while providing comprehensive analytics capabilities.
3Productivity
If deep learning accelerators are integrated into surveillance cameras, then productivity and measurement precision improve, but manufacturing precision and device complexity worsen due to integration challenges
Solution Approach 1:
The removable media serves as an intermediary that bridges the gap between simple surveillance cameras and complex AI processing requirements. Rather than integrating DLA directly into camera manufacturing, the solution uses a separate removable component that interfaces with the camera through standard connections, dramatically simplifying manufacturing while enabling real-time AI processing.
Solution Approach 2:
The system employs dynamic configuration where the removable media can be added, removed, or upgraded based on specific surveillance needs. The DLA can be programmed with different AI models and algorithms that can be updated without hardware changes, providing flexibility and adapting to evolving requirements while maintaining ease of manufacture.
4Measurement precision
If video data is transmitted to cloud servers for processing, then measurement precision improves, but loss of time and energy consumption increase due to data transmission requirements
Solution Approach 1:
The system performs preliminary AI processing and analytics generation locally using the DLA before any cloud transmission. Video frames are analyzed in real-time, and only the resulting analytics (not the raw video data) are transmitted to cloud servers. This preliminary action eliminates the need for transmitting large volumes of video data, reducing transmission time and energy consumption while maintaining analytics accuracy.
Data Source
AI summary
Systems, devices, and methods related to a deep learning accelerator and memory are described. For example, a removable media (e.g., a memory card, or a USB drive) may be configured to execute instructions with matrix operands and configured with: an interface to receive a video stream; and random access memory to buffer a portion of the video stream as an input to an artificial neural network and to store instructions executable by the deep learning accelerator and matrices of the artificial neural network. Such a removable media can be used to replace an existing removable media used in a surveillance camera to record video or images. The deep learning accelerator can execute the instructions to generate analytics of the buffer portion using the artificial neural network, enabling the surveillance camera that is upgraded via the use of the removable media to provide intelligent services based on the analytics.