Semi-Automated Image Segmentation for ADAS Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The manual segmentation and annotation of images for training vision-related machine learning models used in Automated Driving Systems (ADS/ADAS) is time-consuming and prone to errors, requiring significant human effort and time to accurately identify and classify objects like parking spots.
Innovation Solution
A semi-automated method and system that allows users to select fiduciary points for objects of interest, with the software automatically defining edges and adding bounding boxes or optimized polygons, which can be manually adjusted and confirmed, improving accuracy and efficiency through a user-friendly graphical interface.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual segmentation and annotation is performed by users, then accuracy in identifying and discriminating objects of interest is improved, but time consumption and inefficiency increase significantly
Solution Approach 1:
The system performs preliminary automated segmentation to generate initial annotations before user review. This preliminary action creates a draft version of the segmentation that users can then refine, rather than requiring users to create annotations from scratch. The automated pre-segmentation handles the initial object identification and boundary estimation, reducing the overall time required while maintaining accuracy through subsequent user validation.
Solution Approach 2:
The system enables self-service annotation by allowing users to easily correct and refine automated segmentation results through simple interactions. Users can adjust boundaries, add or remove objects, and modify annotations without requiring deep expertise or extensive manual work. This self-service approach combines automated efficiency with human oversight, achieving both speed and accuracy.
2Reliability
If manual segmentation and annotation is performed by users, then robustness in identifying objects of interest is improved, but error proneness and inefficiency increase
Solution Approach 1:
The system introduces an intermediary automated segmentation process that bridges the gap between raw images and final annotations. This intermediary layer performs initial object detection and boundary estimation, which users then review and refine. The intermediary automated process handles routine identification tasks, allowing users to focus on edge cases and complex scenarios, thereby improving both robustness and efficiency.
Solution Approach 2:
The system replaces the purely mechanical manual annotation process with a hybrid approach that uses automated image processing algorithms. These algorithms substitute for human eyes and hands in performing initial segmentation tasks, freeing users to focus on quality control and complex decision-making. This substitution maintains robustness through human oversight while dramatically improving productivity.
3Loss of time
If automated methods are used for segmentation and annotation, then time consumption is reduced, but accuracy and precision in defining edges and bounding boxes deteriorate
Solution Approach 1:
The automated system performs preliminary segmentation to generate initial annotations quickly, providing a starting point for user refinement. This preliminary automated action handles the time-consuming aspect of processing large numbers of images, while the subsequent user review ensures precision in defining edges and bounding boxes. The two-stage process leverages the speed of automation and the precision of human judgment.
Solution Approach 2:
The system incorporates feedback loops where user corrections to automated segmentation are fed back to improve future automated processing. User interactions with the automated results provide training data that refines the automated algorithms over time, gradually improving precision while maintaining time efficiency. This feedback mechanism allows the system to learn from human expertise and reduce the gap between automated speed and human precision.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Methods and systems (10, 40) for digital image segmentation and annotation, including: receiving a digital image (14) depicting, in part, an object of interest from an input file (12, 32); one or more of manually and automatically adding a polygon around the object of interest to generate a segmented digital image (18, 20, 38, 40); one or more of manually and automatically appending a label to the polygon around the object of interest to generate a segmented and annotated digital image (24, 44), wherein the label indicates one or more of an identity and a characteristic of the object of interest; and outputting information related to the segmented and annotated digital image to an output file (26, 48). Optionally, the polygon is one of a bounding box and a 4-point polygon. Optionally, the object of interest is a parking spot.