Method and device for learning a depth prediction model associated with a multi-resolution vision system
A convolutional neural network-based method for ADAS systems optimizes the learning of depth prediction models using stereoscopic vision systems with two cameras, addressing efficiency and reliability challenges by processing common field-of-view pixels and enhancing accuracy in textureless areas.
FR3162893A1Pending Publication Date: 2025-12-05STELLANTIS AUTO SAS
View PDF 0 Cites 0 Cited by
Patent Information
- Application Number
- FR2024005506
- Authority / Receiving Office
- FR · FR
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2024-05-29
- Publication Date
- 2025-12-05
Smart Images
Figure 00000000_0000_ABST
Abstract
A method or device for determining the depth of a pixel in an image using a depth prediction model implemented by a convolutional neural network associated with a vision system comprising several cameras. Images acquired by the vision system are resized (34) and multiplied for different resolutions (35). The depth prediction model is learned by comparing images of different resolutions to generated images (37) for each resolution and predicted depths (36) for pixels in these images, allowing the determination of a first and second error for each pair of images at a given resolution. The depth prediction model is learned (39) by minimizing an error determined from the first and second errors. Figure for the abstract: Figure 3
Need to check novelty before this filing date? Find Prior Art