Vehicle Object Detection Retraining for Low-Confidence Trailer Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current vehicle systems face challenges in accurately identifying and classifying objects, particularly trailers, in images collected by vehicle sensors, especially when the confidence level of identification is low, which can impact safe operation, especially when reversing.
Innovation Solution
A system utilizing machine learning programs, including deep learning algorithms, is trained with a dataset of images annotated with object labels to identify object parameters such as trailer angles, with confidence levels based on standard deviation, and includes a process to retrain the program when high confidence images are received, improving identification accuracy and enabling vehicle components to actuate accordingly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If a machine learning program is used to identify objects in images, then object detection can be automated, but the confidence level of identification may be low leading to inaccurate detection
Solution Approach 1:
The system implements feedback by using high-confidence detections to retrain the machine learning program, improving future detection accuracy. The program continuously learns from its own high-confidence predictions, creating a self-improving loop that enhances reliability while maintaining automation.
Solution Approach 2:
The system performs preliminary classification by a second machine learning program to identify candidate objects before the first program processes them. This preliminary action filters and prioritizes images, allowing the main detection program to focus computational resources on promising candidates, thereby improving overall confidence levels.
2Measurement precision
If the machine learning program processes all images to ensure accurate identification, then detection accuracy improves, but processing time and computational resources increase
Solution Approach 1:
The system applies partial action by processing only a subset of images through the full detection pipeline. The second machine learning program pre-filters images to identify only those containing potential objects of interest, so the first program processes fewer images, reducing total processing time while maintaining accuracy for relevant detections.
Solution Approach 2:
The detection process is segmented into two distinct stages: a preliminary classification stage by the second machine learning program that quickly identifies candidate images, and a detailed detection stage by the first program that processes only those candidates. This segmentation divides the workload to optimize both speed and accuracy.
3Measurement precision
If the training dataset is continuously expanded with more images, then identification accuracy improves, but data processing and model retraining time increase
Solution Approach 1:
The system performs preliminary selection by the second machine learning program to identify high-quality training images with high confidence detections before adding them to the training dataset. This preliminary filtering ensures that only the most valuable data is processed and added, reducing the volume of data requiring retraining while maximizing accuracy improvements.
Solution Approach 2:
The system changes the parameter of confidence level thresholds dynamically. By adjusting the confidence threshold for selecting images to add to the training set, the system can control the quality and quantity of training data, balancing accuracy improvement against retraining time investment.
Data Source
AI summary
A computer is programmed to determine a training dataset that includes a plurality of images each including a first object and an object label, train a first machine learning program to identify first object parameters of the first objects in the plurality of images based on the object labels and a confidence level based on a standard deviation of a distribution of a plurality of identifications of the first object parameters, receive, from a second machine learning program, a plurality of second images each including a second object identified with a low confidence level, process the plurality of second images with the first machine learning program to identify the second object parameters with a corresponding second confidence level that is greater than a second confidence level, retrain the first machine learning program based on the identified second object parameters.