Bounding Box Geometry Training for Yaw-Accurate Object Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous vehicle systems face challenges in accurately generating bounding boxes for object detection using sensor data, particularly in determining the geometry and orientation of objects, which affects navigation and safety.
Innovation Solution
The system employs a machine-learned model that calculates yaw error and intersection over union losses between candidate bounding boxes and ground truth boxes, aligning candidate boxes to improve geometric accuracy and reduce computational complexity, thereby enhancing the training process and object detection confidence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional object detection methods are used to generate bounding boxes from sensor data, then the system can detect objects in the environment, but the geometric accuracy and orientation precision of the bounding boxes are insufficient, affecting navigation safety
Solution Approach 1:
The patent implements a feedback mechanism by calculating IoU loss and yaw error loss between predicted bounding boxes and ground truth bounding boxes during training. The model uses these loss values to iteratively adjust its parameters, improving geometric accuracy over multiple training cycles until the predicted bounding boxes closely match the ground truth in both position and orientation.
Solution Approach 2:
The patent performs preliminary alignment of candidate bounding boxes by calculating yaw error and applying rotational transformations before final detection. This preliminary action of pre-aligning boxes based on predicted yaw angles improves the geometric accuracy of subsequent detection results and reduces computational complexity in the main detection pipeline.
2Measurement precision
If the system calculates detailed geometric losses for training, then the model accuracy improves, but the computational complexity increases
Solution Approach 1:
The patent segments the geometric loss calculation into two independent components: IoU loss for position accuracy and yaw error loss for orientation accuracy. This segmentation allows the model to train on each geometric aspect separately using efficient computational methods, reducing the overall computational complexity compared to calculating a single complex geometric loss.
Solution Approach 2:
The patent transforms the bounding box representation by introducing yaw angle as an explicit parameter and using polar coordinate transformations for loss calculation. This parameter change enables more efficient computation of geometric losses by working in a coordinate system that naturally separates position and orientation components, reducing computational complexity while maintaining detection accuracy.
Data Source
AI summary
Techniques for training a model for detecting objects in an environment are discussed herein. For example, techniques can include determining losses associated with spatial features of candidate bounding boxes output by a machine-learned (ML) model and utilizing the losses to train the ML model. Techniques may include determining candidate bounding box(es) associated with an object detected in an environment using the ML model and receiving a ground truth bounding box associated with the detected object. A yaw error loss may be determined by comparing yaw features of the candidate bounding box to the ground truth bounding box. The candidate bounding box may be axis aligned with respect to the ground truth bounding box and an intersection over union (IoU) loss may be determined based on an IoU between the axis aligned candidate bounding box and the ground truth bounding box. The ML model may be trained based on the losses.