Edge Video Object Detection via Segmented Deep Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current security systems using edge computing networks face challenges in efficiently processing video data from security cameras due to limited computing resources, leading to high bandwidth consumption and the need for costly hardware upgrades.
Innovation Solution
A method and system that utilize a processor and server combination in an edge computing network to detect movement and classify objects of interest using deep neural network algorithms, with a dynamic confidence threshold to reduce processing resources and false positives, allowing for accurate event detection without hardware upgrades.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning algorithms are used to detect objects and events in video surveillance feeds, then detection accuracy is improved, but computing resources required increase significantly
Solution Approach 1:
The patent segments the video processing task into two distinct stages: a first machine learning algorithm runs on edge devices (processors coupled to security cameras) to perform initial object detection and filtering, while a second machine learning algorithm runs on cloud computers to perform more complex analysis on selected video segments. This segmentation allows accurate detection to be achieved while distributing computing resources across multiple locations, reducing the burden on any single device.
Solution Approach 2:
The system applies partial action by having the first machine learning algorithm process only specific regions or frames of video data that contain detected objects, rather than analyzing entire video streams continuously. The second algorithm then applies excessive action by performing comprehensive analysis on these already-filtered segments, achieving high detection accuracy without the need to process all video data at maximum computational intensity.
2Power
If additional hardware devices such as GPUs are added to enhance computing resources, then processing capability is improved, but cost and space requirements increase
Solution Approach 1:
The patent introduces cloud computers as an intermediary between the edge devices and the video data source. The processors coupled to security cameras (edge devices) perform initial processing and send selected data to cloud computers for further analysis. This intermediary approach allows the system to achieve enhanced processing capability without requiring every local device to have expensive hardware upgrades like additional GPUs, as the cloud infrastructure provides the additional computational power.
Solution Approach 2:
The system uses software-based machine learning algorithms that can be deployed and copied across multiple devices (edge devices and cloud computers) without requiring physical hardware duplication. The same detection models and algorithms can be instantiated on different hardware platforms, allowing processing capability to be enhanced through software deployment rather than physical hardware expansion.
Data Source
AI summary
Methods and systems for identifying events of interest in images from imagers in an edge computing network, including: detecting a movement based on image processing in images captured by the imager, cropping an area of the at least one image where the movement is detected, applying a first machine learning (ML) algorithm having a deep neural network architecture to classify an object of interest in the cropped area of the at least one image, sending the at least one image to a server, when the first ML algorithm classifies the object of interest, applying a second ML algorithm having a deep neural network architecture to detect the object of interest in the at least one image where the object of interest is detected based on context analysis.