Image Labeling Overlay With Segmentation for Faster Vehicle Data Annotation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The manual generation of labelling data for training artificial neural networks in autonomously driving vehicles is time-consuming due to the need for hundreds of hours of video material and requires operators to manually describe objects in images, which is inefficient.
Innovation Solution
A processing device generates labelling data by using a segmentation unit to detect objects in images, creating marker data to highlight objects, and receiving user inputs to associate semantic descriptions with these markers, thereby speeding up the labelling process through augmented reality and voice or gesture recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labelling is performed by operators watching image data in a laboratory, then labelling accuracy is improved, but time consumption increases significantly
Solution Approach 1:
The system performs preliminary object detection and segmentation automatically before manual labelling, pre-processing the image data to identify and segment objects of interest. This preliminary action reduces the time required for manual labelling while maintaining accuracy, as operators only need to verify and refine pre-segmented regions rather than manually analysing entire images.
Solution Approach 2:
The patent introduces an intermediary automated object detection and segmentation system that bridges between raw image data and final manual labelling. This intermediary process generates pre-segmented object regions that assist operators, reducing their workload and time consumption while preserving labelling accuracy through human verification.
2Quantity of substance
If hundreds of hours of video material are collected for training data, then training data quantity is improved, but data processing time increases
Solution Approach 1:
The system extracts only the essential objects and regions of interest from hundreds of hours of video material using automated detection and segmentation. Instead of processing entire video datasets manually, the system extracts relevant object instances and their bounding boxes, significantly reducing the volume of data requiring manual labelling while maintaining sufficient training data quantity.
Solution Approach 2:
The patent uses automated object detection to generate copies or representations of objects in video frames, creating structured data (bounding boxes, segmentation masks) that can be directly used for training. This copying process transforms unstructured video data into structured training data efficiently, reducing processing time while maintaining data quantity.
Data Source
AI summary
A method for generating labeling data is disclosed that describes an image content of images depicting at least one scene, wherein in a processing device image data are received from an imaging and a segmentation unit that detects at least one object in the image data. A graphical processing unit generates a respective graphical object marker that marks the at least one detected object and a display control unit displays an overlay of the at least one scene and the at least one object marker. An input reception unit receives a respective user input for each object marker, wherein the respective user input provides the image content of the image region marked by the object marker.