Object Detection Confidence Scoring for Adjacent Vehicle Targets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional object detection systems for autonomous vehicles face inaccuracies due to difficulties in assigning objects to single spatial element regions, especially when objects are close or occupy multiple regions, leading to false or missed detections.
Innovation Solution
A Machine Learning Model, such as a multi-layer perceptron network, is trained to determine a confidence score using features from detected objects, allowing objects to be assigned to multiple spatial element regions and using soft coverage values, and a dead-zone area is employed to separate coverage values, improving the distinction between adjacent objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional CNN approaches are used to detect objects in spatial element regions, then the system can provide coverage values and bounding boxes for real-time detection, but the accuracy decreases leading to false or missed detections
Solution Approach 1:
The patent divides the image into multiple spatial element regions (grid cells) and processes each region independently through the CNN. This segmentation allows the system to handle objects at different locations simultaneously while maintaining dedicated detection capabilities for each region, improving both reliability and measurement precision.
Solution Approach 2:
The patent introduces an intermediary confidence determination step between object detection and final output. Instead of directly using coverage values as detection results, the system determines confidence values based on multiple factors including coverage, spatial position, and object characteristics. This intermediary layer filters false detections and improves detection accuracy.
2Reliability
If objects are assigned to multiple spatial element regions, then the system can reduce missed detections, but the complexity of assigning and aggregating detections increases
Solution Approach 1:
The patent merges detections from multiple spatial element regions by aggregating coverage values and bounding boxes for objects that span across region boundaries. The system combines detection results from adjacent regions using weighted averaging or maximum selection, reducing missed detections while managing complexity through systematic aggregation rules.
Solution Approach 2:
The patent handles multi-region object assignment by introducing an additional dimension of confidence scoring. Instead of simply aggregating detections, the system evaluates confidence values that incorporate spatial relationships, coverage proportions, and object continuity across regions. This additional dimensional approach manages the complexity of multi-region assignments.
3Ease of operation
If maximum coverage value is used as confidence value, then the system can provide a simple confidence measure, but false detections increase as it only corresponds to a small portion of the image
Solution Approach 1:
The patent applies local quality by determining confidence values differently based on the characteristics of each detection. Instead of using a uniform confidence metric, the system adjusts confidence calculations based on local factors such as coverage proportion, spatial position, object size, and detection consistency. This localized approach improves detection reliability while maintaining operational simplicity through context-aware confidence determination.
Data Source
AI summary
In various examples, detected object data representative of locations of detected objects in a field of view may be determined. One or more clusters of the detected objects may be generated based at least in part on the locations and features of the cluster may be determined for use as inputs to a machine learning model(s). A confidence score, computed by the machine learning model(s) based at least in part on the inputs, may be received, where the confidence score may be representative of a probability that the cluster corresponds to an object depicted at least partially in the field of view. Further examples provide approaches for determining ground truth data for training object detectors, such as for determining coverage values for ground truth objects using associated shapes, and for determining soft coverage values for ground truth objects.