Hybrid Cloud Camera AI Vision System
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current security systems using cloud-connected cameras for AI-driven video processing face high costs and delays due to extensive bandwidth and compute resource usage, while on-device neural networks compromise accuracy for reduced complexity.
Innovation Solution
A hybrid cloud/camera AI computer vision system where smart cameras capture video frames, detect triggering events, and upload selected images to the cloud for analysis, reducing bandwidth and compute load while enabling real-time object detection and feedback.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If constantly stream video to cloud service for neural network processing, then detection accuracy is improved, but bandwidth usage and server compute resources increase significantly
Solution Approach 1:
The video stream is segmented into individual frames, and only selected frames are uploaded to the cloud for processing. The camera performs local frame selection based on motion detection, transmitting only relevant portions of the video data rather than the entire continuous stream, thus reducing bandwidth and compute resource requirements while maintaining detection accuracy.
Solution Approach 2:
The system extracts and transmits only the essential information (selected frames containing motion events) to the cloud service, separating the important data from the redundant continuous video stream. This extraction approach eliminates unnecessary data transmission while preserving the critical detection information needed for accurate object recognition.
2Loss of energy
If start video stream upload after motion trigger is detected, then bandwidth usage and server compute resources are reduced, but detection delay increases by multiple seconds
Solution Approach 1:
The system performs preliminary motion detection and frame selection locally at the camera before uploading to the cloud. By pre-processing and identifying relevant frames in advance, the system reduces the time needed for cloud processing and eliminates delays associated with waiting for motion triggers, achieving both resource efficiency and timely detection.
3Loss of time
If modify or compress neural network to run on embedded circuitry of smart camera, then cloud server cost and transmission delay are eliminated, but predictive power and accuracy of detections are reduced
Solution Approach 1:
The neural network processing is segmented between the camera and cloud service. The camera performs lightweight local processing for motion detection and frame selection, while the cloud service handles the computationally intensive neural network analysis for accurate object recognition. This segmentation allows each component to operate at its optimal capability level.
Solution Approach 2:
The system uses an intermediary approach where the camera selects and transmits representative frames to the cloud, which then performs comprehensive neural network analysis. This intermediary frame selection mechanism enables the system to leverage both the speed of on-device processing and the accuracy of cloud-based neural networks without requiring the full neural network to run locally.
Data Source
AI summary
A system including a smart camera and a server computer. The smart camera may be configured to capture video data comprising a plurality of video frames, detect a triggering event and select a plurality of images from the plurality of video frames of the video data in response to the triggering event and communicate the plurality of images. The server computer may be configured to receive the plurality of images, analyze the plurality of images to perform an additional detection corresponding to the triggering event and generate a signal in response to the additional detection. The server computer may implement a neural network for analyzing the plurality of images. The neural network may use more computing resources than the smart camera uses for detecting the triggering event. Communicating the plurality of images may use less bandwidth than communicating the video data.