Weighted Score Fusion in Multi-Recognizer Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection devices face accuracy issues due to varying recognition performance among multiple recognizers, leading to potential degradation of final results and increased device size, making them unsuitable for deployment in video monitoring or similar applications.
Innovation Solution
An object detection device comprising multiple object detection units that output scores for partial regions, a weight computation unit to merge these scores using weight computation parameters, and a loss computation unit to correct these parameters for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple recognizers with different characteristics are constructed to improve recognition accuracy, then the recognition accuracy is improved, but the device becomes large scale and complex
Solution Approach 1:
The patent combines multiple recognizers with different characteristics into a unified object detection device. The merging unit integrates the determination results from multiple recognizers, allowing the system to leverage the strengths of each recognizer while maintaining a manageable device scale. This resolves the contradiction by achieving improved recognition accuracy through combination without proportionally increasing device complexity.
Solution Approach 2:
The object detection device is designed to perform multiple functions by incorporating recognizers with different characteristics. Each recognizer can handle different types of objects or detection scenarios, making the device universally applicable to various detection tasks. This multi-functionality approach allows the device to maintain versatility while avoiding the need for separate specialized devices for each detection scenario.
2Measurement precision
If multiple recognizers are used to improve recognition performance, then the recognition accuracy is improved, but the device becomes large scale making it unsuitable for site deployment
Solution Approach 1:
The patent merges multiple recognizers into a single integrated object detection device that can be deployed at sites. The merging unit combines the outputs of multiple recognizers in a way that maintains compact device architecture, making the system suitable for on-site deployment in video monitoring applications while preserving the accuracy benefits of multiple recognizers.
Solution Approach 2:
The merging unit acts as an intermediary that coordinates the outputs of multiple recognizers. This intermediary component manages the integration process efficiently, allowing the system to maintain a compact and deployable structure while still utilizing multiple recognizers for improved accuracy. The merging unit mediates between the multiple recognizers and the final output, enabling practical site deployment.
3Measurement precision
If multiple determination units are used to make final determination, then the recognition accuracy is improved, but the device complexity increases
Solution Approach 1:
The patent combines multiple determination units into a unified decision-making process. The merging unit integrates the determination results from multiple recognizers using a standardized approach, reducing the complexity that would otherwise arise from managing multiple independent determination systems. This allows the system to achieve improved final determination accuracy while maintaining manageable system complexity.
Data Source
AI summary
In an object detection device, the plurality of object detection units output a score indicating a probability that a predetermined object exists for each partial region set with respect to inputted image data. The weight computation unit uses weight computation parameters to compute a weight for each of the plurality of object detection units on a basis of the image data and outputs of the plurality of object detection units, the weight being used when the scores outputted by the plurality of object detection units are merged. The merging unit merges the scores outputted by the plurality of object detection units for each partial region according to the weights computed by the weight computation unit. The first loss computation unit computes a difference between a ground truth label of the image data and the score merged by the merging unit as a first loss. Then, the first parameter correction unit corrects the weight computation parameters so as to reduce the first loss.