Vehicle Object Detection Using Hierarchical Attribute Confidence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge lies in accurately detecting objects, especially rare ones, in images with poor quality due to factors like glare or motion blur, where insufficient training data and unclear features lead to undetected objects, impacting automated driving systems.
Innovation Solution
An object detection device employing a multi-tiered neural network system that separately identifies position regions, large attributes, and small attributes, with an object judging part that determines the detection result based on confidence thresholds and attribute hierarchies, ensuring objects are not missed even in difficult detection scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single neural network is used for object detection, then the device complexity is low, but the detection accuracy for rare objects and objects in poor quality images deteriorates
Solution Approach 1:
The detection system is segmented into multiple specialized neural networks: a first neural network for position region detection, a second neural network for large attribute identification, and a third neural network for small attribute identification. Each network focuses on specific detection tasks, improving overall accuracy while managing complexity through functional division.
Solution Approach 2:
The system adds a hierarchical dimension to detection by introducing attribute identification layers (large attributes and small attributes) beyond simple object detection. This multi-dimensional approach allows the system to detect objects with varying levels of detail, improving accuracy for rare and obscured objects.
2Reliability
If multiple neural networks are used to improve detection accuracy, then the detection reliability improves, but the processing time and computational resources increase
Solution Approach 1:
The first neural network performs preliminary detection of position regions before the attribute identification networks are engaged. This preliminary action filters and prepares data, reducing the computational burden on subsequent networks and minimizing overall processing time while maintaining reliability.
Solution Approach 2:
The system dynamically adjusts the detection process based on confidence levels. When the first neural network achieves high confidence in position detection, the system can reduce the depth of attribute analysis, creating a dynamic processing path that balances reliability and speed based on actual detection conditions.
3Measurement precision
If a strict confidence threshold is applied for object detection, then the false positive rate decreases, but the number of undetected objects increases
Solution Approach 1:
The system addresses the threshold dilemma by adding attribute identification dimensions. Even when position detection confidence is below the strict threshold, the attribute identification networks can provide additional evidence through large and small attribute analysis, allowing the system to maintain high precision while reducing false negatives through multi-dimensional verification.
Data Source
AI summary
An object detection device 30 comprises a position region detecting part 31 using a first neural network to detect a position region of an object in the image, a large attribute identification part 32 configured using a second neural network to identify a large attribute of the object, a small attribute identification part 33 using a third neural network to identify a small attribute of the object, and an object judging part 34 judging a result of detection of the object. The object judging part is configured to judge that a result of identification of the small attribute is the result of detection if a confidence of the result of identification of the small attribute is equal to or more than a threshold value, and judge the result of detection based on a result of identification of the large attribute if the confidence is less than the threshold value.