Object Detection Confidence Fusion for Surround View Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The confidence of object detection results in surround view images is excessively low, leading to uncertainty about the reality of detected objects and increased error rates.
Innovation Solution
Fusing the confidence levels of objects detected from single images and surround view images, using weighted averaging based on threshold comparisons and image coincidence, to determine a fusion confidence for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If object detection is performed on surround view image, then detection coverage is improved, but detection accuracy deteriorates due to excessively low confidence
Solution Approach 1:
The patent merges confidence information from multiple detection sources (single image detection and surround view image detection) to create a fusion confidence value. This combining approach allows the system to maintain the broad detection coverage of surround view images while improving detection accuracy by incorporating confidence metrics from both detection methods.
Solution Approach 2:
The patent implements a feedback mechanism where the confidence values from single image detection are used to adjust and refine the confidence assessment of objects detected in surround view images. This feedback loop enables the system to correct low confidence values in surround view detections by referencing the confidence information from more reliable single image detections.
2Quantity of substance
If confidence threshold is lowered to detect more objects, then detection completeness is improved, but false positive rate increases
Solution Approach 1:
The patent creates a composite confidence metric by combining confidence values from two different detection sources. This composite confidence value allows the system to maintain a high detection threshold while still detecting a large number of objects, because the fusion of multiple confidence assessments provides more reliable indication of true objects versus false positives.
Data Source
Figure 1~2
Figure 3~4B
Figure 5~7
AI summary
This application belongs to the field of sensor technologies, and discloses an object detection method and an apparatus, to ensure a trustworthiness level of a detected object and further improve accuracy of object detection. The object detection method includes: when a first object and a second object are a same object, obtaining a fusion confidence based on a first confidence and a second confidence, where the first confidence is a trustworthiness level of the first object in a first image, the second confidence is a trustworthiness level of the second object in a second image, the first image is an image from a first image shooting apparatus, the second image is a surround view image obtained based on images from a plurality of image shooting apparatuses, and the plurality of image shooting apparatuses include the first image shooting apparatus; and outputting an object detection result based on the fusion confidence. The method and the apparatus may be applied to object detection and tracking in assisted driving and autonomous driving.