ROI-Based Target Detection for Small Object Misidentification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial intelligence technologies struggle to accurately identify small targets in images, particularly in scenarios like autonomous driving, where distant or small objects are misidentified, such as mistaking a distant tree for a pedestrian.
Innovation Solution
A target detection method that involves obtaining an image using a photographing apparatus, marking a region of interest based on the apparatus' parameters and a preset path, and using a target detection algorithm to determine the target's category and location, with confidence adjustments based on the relative location relationship and error analysis between the detected region and the region of interest.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If a target detection algorithm is used to detect small targets in images, then detection coverage is improved, but detection accuracy deteriorates due to misidentification of distant or small objects
Solution Approach 1:
The patent implements a feedback mechanism by calculating confidence levels for detected targets and adjusting detection results based on whether confidence exceeds a threshold. The system uses the relationship between detected target location and region of interest to modify confidence, creating a closed-loop verification process that improves accuracy while maintaining coverage
Solution Approach 2:
The patent introduces confidence level as an intermediary parameter between target detection and final identification. This mediator allows the system to evaluate detection reliability and filter out uncertain detections, resolving the contradiction between detecting all potential targets and ensuring accurate identification
2Measurement precision
If confidence modification is performed based on location relationship, then detection accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent applies local quality by performing confidence modification only for detected targets that fall within or near the region of interest. Instead of processing all detected objects uniformly, the system selectively applies additional verification steps only where needed, reducing overall computational complexity while maintaining accuracy for critical detections
Solution Approach 2:
The patent performs preliminary marking of the region of interest based on photographing apparatus parameters and preset traveling path before target detection. This preliminary action establishes a reference framework that simplifies subsequent confidence calculation and reduces the computational burden during the main detection phase
Data Source
AI summary
Embodiments of this application provide example target detection methods and apparatuses. One target detection method includes obtaining an image by using a photographing apparatus. A region of interest can be marked in the image based on a parameter of the photographing apparatus and a preset traveling path. The image can be detected by using a target detection algorithm to obtain a category to which a target object in the image belongs, a first location region of the target object in the image, and a confidence of the category. The confidence of the category can be modified, based on a relative location relationship between the first location region and the region of interest, to obtain a first confidence.