Driving Object Detector Training With Size-Based Vehicle Anchors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection systems for autonomous vehicles and ADAS face inaccuracies due to error-prone reference information and resource-intensive adjustments, particularly in distinguishing between insignificant differences in object sizes, leading to suboptimal performance and high computational costs.
Innovation Solution
A method utilizing a shallow neural network with a region unit, including dedicated sections for four-wheel vehicles, two-wheel vehicles, and pedestrians, employs unique anchor sets and size-based classification to improve accuracy and reduce resource consumption by distinguishing significant differences, and is trained to detect objects within specific size ranges to prevent errors in object detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the object detector is adjusted to compensate for insignificant differences between objects, then the accuracy of object detection is improved, but the computational resources consumed increase significantly
Solution Approach 1:
The patent segments the object detection task by dividing objects into different size ranges (first size range for four-wheel vehicles, second size range for subclasses). This segmentation allows the system to apply different detection strategies to different object categories, avoiding the need to adjust for all possible insignificant differences across all object types, thereby reducing computational resources while maintaining accuracy for significant distinctions.
Solution Approach 2:
The patent applies local quality by assigning unique anchor sets to different object classes and size ranges. Each object category receives tailored detection parameters appropriate to its characteristics, rather than applying uniform adjustments to all objects. This localized approach improves detection accuracy for each category while avoiding unnecessary computational adjustments for insignificant differences.
2Ease of manufacture
If reference information is generated manually or by other error prone processes, then the object detector can be trained, but the accuracy of the error calculation is reduced
Solution Approach 1:
The patent changes the parameter of object classification by introducing size-based categorization (first size range vs. second size range) and unique anchor sets for different classes. This parameter change transforms the training process into a more structured approach where objects are systematically categorized before detection, reducing errors in reference information generation and improving error calculation accuracy while maintaining training feasibility.
3Device complexity
If the object detector detects all subclasses of four-wheel vehicles with the same detection process, then the device complexity is reduced, but the detection accuracy for specific vehicle types decreases
Solution Approach 1:
The patent segments four-wheel vehicles into different size ranges, with the first size range detected as general four-wheel vehicles and the second size range detected as specific subclasses (cars, trucks, buses, vans). This segmentation maintains relatively simple device complexity by using size-based categorization while improving subclass detection accuracy through dedicated anchor sets and detection parameters for each category.
Data Source
AI summary
A method for driving-related object detection, the method may include receiving an input image by an input of an object detector; and detecting, by an object detector, objects that appear in the input image. The detecting includes searching for (i) a first object having a first size that is within a first size range and belongs to a four wheel vehicle class, (ii) a second object having a second size that is within a second size range and belongs to a subclass out of multiple four wheel vehicle subclasses, (iii) a pedestrian, and (iv) a two wheel vehicle; wherein a maximum of the first size range does not substantially exceed a minimum of the second size range.