Distributed Camera Inference for Low-Bandwidth Item Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Surveillance systems face challenges in efficiently searching for and locating items of interest due to the high communication bandwidth and computing power requirements for processing large volumes of video data from distributed security cameras.
Innovation Solution
A distributed camera system is implemented with local Deep Learning Accelerators (DLAs) that process video images using artificial neural networks, reducing the need for centralized computing and communication by classifying relevant images locally and transmitting only matching results to a central station.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If video images from distributed security cameras are transmitted to a central station for processing, then centralized computing can identify items of interest, but communication bandwidth requirements and central processing load increase significantly
Solution Approach 1:
The system divides the centralized processing task into distributed local processing units at each camera. Each camera's Deep Learning Accelerator independently processes video frames using the item model, segmenting the overall computation across multiple spatial locations. This reduces the need to transmit all video data to a central station, thereby decreasing communication bandwidth requirements while maintaining identification accuracy through distributed inference.
2Measurement precision
If all video data is processed at a central station, then comprehensive analysis can be performed, but processing time and system response delay increase
Solution Approach 1:
The system performs preliminary processing actions at each distributed camera by executing the item model inference locally before any potential central review. The Deep Learning Accelerators at each camera pre-process video frames in real-time, identifying items of interest locally. This preliminary local action eliminates the need to wait for centralized processing of all data, significantly reducing response delay while maintaining comprehensive analysis capabilities through the distributed network.
3Quantity of substance
If Deep Learning Accelerators are deployed at each camera, then local processing reduces communication bandwidth, but device complexity at each camera increases
Solution Approach 1:
The system replaces the mechanical/physical transmission of large volumes of video data through communication networks with computational processing at the source. Instead of mechanically transmitting all video frames over bandwidth-constrained channels, Deep Learning Accelerators perform computational inference locally at each camera. This substitution of data transmission with local computation reduces communication bandwidth requirements, and the complexity is managed through standardized integrated circuit designs that can be deployed uniformly across the distributed camera network.
Data Source
AI summary
A system having a central station and a plurality of cameras installed various locations. To search for and locate an item of interest, the central station generates and sends an item model to the cameras. When stored in a camera, the item model causes a logic circuit of the camera (e.g., a deep learning accelerator) to use image data, received from an image sensor for storing in a memory device of the camera, as an input to an artificial neural network. The logic circuit performs the matrix computation of the artificial neural network to generate a classification of whether the images are relevant to the item of interest characterized by the item model. If so, the camera transmits the relevant images to the central station for further processing to determine a real time location of the item of interest.