Hash Vector Scoring for Video Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video surveillance systems require significant computing resources for accurate object detection and classification in video footage, making it time-consuming to identify specific objects, such as a lost child in a large area, and there is a need to optimize resource utilization for efficient processing.
Innovation Solution
A method using a processor to generate a hash vector from image data, determine which components satisfy a threshold criterion, and create a scoring database to assess similarity with target examples, allowing for efficient object detection and classification by distributing processing tasks and utilizing hash-weight pairs for scoring.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If computer-implemented analysis of video is performed to detect and recognize objects with high accuracy, then object detection accuracy is improved, but computing resources required increase substantially
Solution Approach 1:
The patent segments the video processing task into multiple independent hash computations, where each hash function processes a different aspect of the visual data. This allows parallel processing across multiple processors or computing nodes, distributing the computational load while maintaining detection accuracy through the collective results of multiple hash comparisons.
Solution Approach 2:
The patent introduces hash values as an intermediary representation between the original visual data and the final object identification. By converting visual features into compact hash codes, the system reduces the computational complexity of direct image comparison while preserving the ability to distinguish between different objects, thereby reducing computing resources required.
2Productivity
If computer-implemented detection of objects is performed to facilitate timely review of video segments, then object detection speed is improved, but computing resources required increase
Solution Approach 1:
The patent applies partial action by using a limited set of hash functions rather than exhaustive image analysis. The system computes only the necessary hash values required for rapid object identification, avoiding full-frame processing while still achieving timely detection. This selective approach maintains productivity while reducing the excessive computational resources that would be required for complete video frame analysis.
3Use of energy by moving object
If manual review of video footage is performed to identify specific objects, then resource utilization is optimized, but time required increases significantly
Solution Approach 1:
The patent performs preliminary action by pre-computing and storing hash values for objects of interest before the actual search begins. When a search is initiated, the system quickly compares the query hash against the pre-computed database, eliminating the need for time-consuming manual review or real-time full-video analysis. This preliminary preparation enables rapid identification while maintaining efficient resource utilization during the search operation.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
Methods, systems, and techniques for performing a hash-based appearance search. A processor is used to obtain a hash vector that represents a search subject that is depicted in an image. The hash vector includes one or more hashes as a respective one or more components of the hash vector. The processor determines which one or more of the hashes satisfy a threshold criterion and which one or more of the components of the hash vector qualify as a scoring component. The one or more components that qualify correspond to a respective one or more hashes that satisfy the threshold criterion and that are represented in a scoring database that is generated based on different examples of a search target. The processor determines a score representing a similarity of the search subject to the different examples of the search target.