Surveillance Video Coding with Background and Foreground Databases
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video surveillance systems face inefficiencies in decoding and recognizing surveillance videos, leading to distorted high-frequency information loss and increased storage and monitoring workload, with manual operations and low accuracy in unfamiliar environments.
Innovation Solution
A method and system that establish background and foreground object databases to code surveillance images, allowing direct analysis and extraction of foreground objects without full decoding, using semantic descriptions for improved monitoring accuracy and reduced costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If surveillance video is compressed and decoded for monitoring, then the video can be displayed and monitored, but high frequency information is lost and the video becomes distorted
Solution Approach 1:
The patent segments the video processing into two distinct paths: a compression path for storage (producing coded video streams) and an analysis path for recognition (using raw video data). By separating these functions, the system avoids decoding compressed video for analysis purposes, thereby preserving high frequency information while still enabling monitoring through the coded streams when needed.
Solution Approach 2:
The system performs preliminary analysis on raw video data before compression occurs. Feature extraction and object recognition are conducted on the uncompressed video streams, allowing the system to identify and track important objects and events. This preliminary action ensures that high frequency information is captured and analyzed before any compression-induced distortion occurs.
2Measurement precision
If full decoding is performed on surveillance video for recognition, then accurate recognition can be achieved, but time is wasted and processing efficiency decreases
Solution Approach 1:
The patent divides video processing into separate functional streams: one for compression/storage and another for real-time analysis. The analysis stream processes raw video data directly without requiring full decoding, enabling parallel processing of multiple video channels simultaneously. This segmentation eliminates the time-consuming decoding step while maintaining recognition accuracy through direct processing of uncompressed data.
Solution Approach 2:
The system performs feature extraction and object recognition as preliminary actions on raw video data before compression is applied. By identifying and tagging important objects, events, and patterns in advance, the system creates metadata that can be quickly queried and analyzed without requiring full video decoding, thus saving significant processing time while maintaining recognition accuracy.
3Adaptability or versatility
If video clips are used as units for analysis in unfamiliar environments, then universal property is achieved, but the recognition process becomes very difficult
Solution Approach 1:
The system performs preliminary learning and adaptation by analyzing raw video data from the specific monitoring environment before full deployment. It identifies environment-specific objects, patterns, and characteristics, building customized recognition models tailored to that environment. This preliminary action allows the system to achieve universal adaptability while reducing recognition difficulty by pre-adjusting to environment-specific features.
Solution Approach 2:
The system incorporates feedback mechanisms where recognition results are continuously refined based on actual video data from the monitoring environment. By comparing expected patterns with actual observations and adjusting recognition parameters accordingly, the system adapts to unfamiliar environments over time, reducing recognition difficulty while maintaining universal applicability across different settings.
4Reliability
If round-the-clock video surveillance is implemented, then complete monitoring coverage is achieved, but storage space is occupied and monitoring workload increases
Solution Approach 1:
The patent extracts and stores only the essential information from continuous video surveillance in the form of coded representations and extracted features. Instead of storing all raw video data, the system extracts key objects, events, and patterns, storing them in a compressed format that occupies minimal storage space while maintaining complete monitoring coverage through the ability to reconstruct or query the extracted information as needed.
Solution Approach 2:
The system discards redundant and unnecessary video data while recovering and preserving only the critical information needed for monitoring and analysis. By continuously filtering and selecting only meaningful content (such as detected objects, unusual events, or changes in monitored areas), the system maintains complete monitoring coverage without accumulating excessive storage requirements, as irrelevant data is discarded and only essential information is retained.
Data Source
AI summary
Method for coding or recognizing of surveillance videos is provided to improve compressing efficiency and recognizing accuracy of surveillance videos. The method for coding surveillance videos includes: establishing a background database and a foreground object database; wherein, the background database includes a set of background images; the foreground object database includes a set of foreground objects; coding a surveillance image by referring to a background image in the background database and a foreground object in the foreground object database.