Autonomous Vehicle Object Classification via Temporal Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional approaches for training artificial intelligence models in autonomous vehicles result in sub-optimal object classification, particularly for objects with occlusion, low resolution, and unusual yaw angles, leading to inaccurate obstacle detection and potential safety risks on public roadways.
Innovation Solution
A method and system that track objects across a sequence of images using bounding boxes and tracking identifiers, apply a voting algorithm to correct classifications, and train the AI model based on these corrections, improving classification accuracy by assigning weight values to images with consistent classifications and determining object distance from the vehicle.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional artificial intelligence models are used for object classification in autonomous vehicles, then the system can operate with standard processing capabilities, but the classification accuracy deteriorates for objects with occlusion, low resolution, and unusual yaw angles
Solution Approach 1:
The system performs preliminary tracking of objects across multiple frames before final classification, establishing temporal context and motion patterns that help disambiguate difficult cases. This preliminary action allows the model to accumulate evidence over time rather than relying on a single ambiguous frame.
Solution Approach 2:
The system implements a feedback mechanism where tracking results inform classification decisions and vice versa. The tracker provides spatial-temporal constraints that feed back into the classification process, allowing the model to correct its own predictions when temporal consistency indicates an error.
2Measurement precision
If the system processes every image in real-time with high computational power, then classification accuracy improves, but the processing time and computational resources increase
Solution Approach 1:
The tracker performs preliminary filtering and candidate selection across image sequences, pre-processing the data in a computationally efficient manner before passing selected candidates to the more intensive classification model. This preliminary action reduces the overall computational burden while maintaining accuracy.
Solution Approach 2:
The system maintains continuous tracking states across frames, preserving useful information from previous frames rather than reprocessing everything from scratch. This continuity allows the system to leverage temporal patterns and reduce redundant computations, improving efficiency without sacrificing accuracy.
Data Source
AI summary
Systems and methods for training artificial intelligence models based on sequences of image data are disclosed. The techniques described herein include generating, using an artificial intelligence model, a respective classification and a respective bounding box for an object depicted in each image of a sequence of images captured during operation of an autonomous vehicle; tracking the object in the sequence of images based on the respective bounding box of each image of the sequence of images and a tracking identifier corresponding to the object; determining a correction to the respective classification of an image of the sequence of images responsive to tracking the object in the sequence of images; and training the artificial intelligence model based on the correction.