AI-supported method for optical processes with video evaluation.
A single-camera system with a positioning unit captures 3D images from multiple angles to address inefficiencies in existing image evaluation methods, enhancing detection accuracy and efficiency by compensating for surface deviations and reducing resource consumption.
Patent Information
- Application Number
- DE102024000375
- Authority / Receiving Office
- DE · DE
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2024-02-05
- Publication Date
- 2025-08-07
AI Technical Summary
Existing image evaluation methods require high time and resource expenditure due to the need for creating reference images and defining test features based on coordinates, are prone to errors from surface deviations and partial concealment, and necessitate multiple cameras for larger specimens, leading to inefficiencies and inaccurate results.
A single video camera mounted on an inspection device with a positioning unit, such as a robot arm, captures a test object from multiple perspectives through dynamic movement, generating high-resolution 3D images and compensating for surface effects, and allows evaluation via video sequences or KI-based processing.
This approach reduces time and resource consumption by using a single camera to capture 3D images from multiple angles, enhancing detection accuracy and compensating for surface deviations, thus improving efficiency and precision.
Abstract
Description
[0001] The invention defined in claim 1 is based on the following problems.
[0002] The usual method for evaluating image content is based on parameters that must be specifically tailored to the static inspection features to be evaluated in the form of image details.
[0003] To do this, a photograph must first be taken with a camera, which serves as a reference image. Using this reference image, the test characteristics are then defined in the form of coordinates in the reference image. For the test, a photograph of the test object is taken that precisely contains the test characteristics of the coordinates of the reference image. The test is evaluated by comparing the test photograph (actual state) with the reference photo (target state). Problem 1: High time expenditure for creating the individual reference photos and defining the test characteristics contained therein based on coordinates. Problem 2: Very high effort when there are many variants of the test object in the form of different colors and positions of the test feature. Problem 3: Deviating surface textures (light scratches, contamination on shiny surfaces) lead to incorrect test results. Problem 4: Partially hidden or difficult to see components of the test object cannot be tested because they are not captured by the static camera. Problem 5: High costs due to the necessary use of several cameras to create test photos for larger test specimens. Problem 6: High time expenditure for the testing of a larger test specimen, since several test photos must be taken and evaluated using several cameras.
[0004] These problems are solved by the features mentioned in patent claim 1.
[0005] At least one positioning unit (e.g., a robot arm) with at least one video camera is mounted on a test fixture. Depending on the type of test piece support at the test station, for example, a conveyor belt or a positionable rotary / swivel table, the camera movement relative to the test piece can be realized in the following ways: 1. Camera moves - test subject stands 2. Camera is stationary - test subject is moving 3. Camera moves - test object moves
[0006] During linear movement, a video sequence is created which records the test object from different perspectives due to the optical path of the lens.
[0007] The evaluation of the created video sequence is characterized by the fact that it a) higher resolution image information is generated from several combined individual video images b) 3D image information is generated from the individual images of the video sequence using the multi-stereo image effect. c) The test subject can be assessed either by means of classical photo evaluation or AI-supported image processing.
[0008] The advantages achieved by the process are characterized by the fact that - only a single camera is required - through multiple viewing angles, both a 3D image and more image information of the test object can be obtained, thus achieving greater recognition accuracy. - Gloss and reflection effects occurring on the surface to be tested can be compensated - through dynamic movement, large volumes of information can be captured in a short time
Claims
[1] Image processing application consisting of: a) a test device b) a video camera c) a positioning unit (e.g. robot arm) d) a reference image e) where the following applies: f) that the reference image contains at least one referenced inspection feature (ROI) During a test, the evaluation software finds the learned ROIs in the test specimen photos. The evaluation is performed by comparing the target and actual values.