| |
This article discusses the application of various AI methods to improve the accuracy and automation of the photogrammetry process to create 3D models. The focus is on using advanced deep learning techniques such as image segmentation, depth mapping, and Image Enhancement (Super-Resolution) to improve key steps of photogrammetric processing. This article analyzes the role of neural networks in automating tasks such as identifying key points and creating accurate bit masks, which allows you to speed up the segmentation process and eliminate errors that occur during manual processing.
The article pays special attention to the two most promising possible integrations of neural network models.
The segmentation method helps to accurately highlight targets in images, separating them from the background and unwanted elements. To demonstrate the application of the approach, the CVAT service is used, based on the pre-trained Mask R-CNN model. The result of the service operation is demonstrated in the form of an image with annotated data in automatic mode, as well as high-quality bit masks created.
Finally, the Super-Resolution method is considered, which allows to increase the resolution of the source images, thereby improving the detail and quality of 3D models. The pre-trained neural network VGG19 is used as a possible tool, and its basic principle of operation is considered. As a result, a model created using improved data is demonstrated. According to the results of the Super-Resolution method, it was possible to achieve a 2.5-fold improvement in the quality of the 3D model.
In conclusion, it is emphasized that the integration of neural networks in photogrammetry opens up the prospect of significantly accelerating and automating processes when it is necessary to prepare a large number of images necessary to create 3D models. The disadvantages are a large amount of training data, and significant time spent on training, as well as very high requirements for computing resources.
Keywords:neural network, Super-Resolution, photogrammetry, image segmentation, bitmask, Depth map
|
LEGAL
INFORMATION:
