Deep Feature Coding for Low-Bandwidth Video Vision Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The large data amount of videos transmitted to the cloud server for vision analysis tasks puts a heavy pressure on data transmission and storage, as existing methods require the entire video to be transmitted and processed.
Innovation Solution
A method for deep feature coding and decoding that extracts features from video frames, determines their types based on time-domain correlation with a reference feature, and encodes them using specific patterns, reducing the data amount transmitted to the cloud server, where the coded features are decoded for analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the entire video is transmitted to the cloud server for vision analysis, then the analysis accuracy is maintained, but the data transmission and storage pressure on the cloud server increases significantly
Solution Approach 1:
The patent extracts only the essential feature information from video frames using deep learning feature extractors, rather than transmitting the entire video data. This extraction process isolates the critical visual features needed for vision analysis tasks while leaving out redundant information, thereby reducing data transmission volume while maintaining analysis accuracy.
Solution Approach 2:
The patent segments the video processing workflow into two parts: local feature extraction and encoding at the terminal device, and cloud-based decoding and analysis at the server. This segmentation allows the computationally intensive feature extraction to be performed locally, reducing the data burden on cloud infrastructure while preserving the analytical capabilities needed for accurate vision analysis.
2Loss of information
If the entire video is transmitted to the cloud server, then complete information is available for analysis, but the storage pressure on the cloud server increases significantly
Solution Approach 1:
The patent extracts only the essential feature information from video frames using deep learning feature extractors, rather than transmitting the entire video data. This extraction process isolates the critical visual features needed for vision analysis tasks while leaving out redundant information, thereby reducing data transmission volume while maintaining analysis accuracy.
Solution Approach 2:
The patent creates compressed representations (coded features) of the original video content that can be stored and transmitted efficiently. These coded features serve as compact copies that retain the essential information needed for vision analysis while occupying minimal storage space on the cloud server.
3Quantity of substance
If deep feature coding is performed at the terminal device, then data transmission volume is reduced, but the device complexity increases
Solution Approach 1:
The patent performs deep learning feature extraction and encoding operations at the terminal device before video transmission, preparing the data in advance. This preliminary processing extracts essential features and encodes them using predetermined coding patterns, reducing the data burden on cloud infrastructure while preserving the analytical capabilities needed for accurate vision analysis.
Solution Approach 2:
The patent introduces an intermediate processing stage at the terminal device that transforms raw video data into coded features before cloud transmission. This intermediary processing layer, implemented through deep learning feature extractors and encoders, acts as a bridge between the video source and cloud analysis systems, reducing data transmission volume while maintaining analysis accuracy.
Data Source
AI summary
Embodiments of the present disclosure provide a method, an apparatus, and a system for deep feature coding and decoding. The method comprises: extracting features of respective video frames; determining types of the features, the types reflecting time-domain correlation degrees between the features and a reference feature; encoding the features using predetermined coding patterns matching the types to obtain coded features; and transmitting the coded features to the server such that the server decodes the coded features for a vision analysis task. By using the embodiments of the present disclosure, videos per se may not be transmitted to the cloud server; instead, the features of the video, after being encoded, are transmitted to the cloud server for a vision analysis task; compared with the prior art, data transmission pressure may be lowered, and the storage pressure at the cloud server may also be lowered.