Visiontek Virtuoso3D automatic tilt photogrammetry 3D modeling cluster system Virtuoso3D

System Overview
Virtuoso3D is a set of high-performance oblique image 3D modeling cluster processing system independently developed by Visiontek. The system is based on the principle of photogrammetry, with the support of intelligent task partitioning and high-performance computing cluster hardware, which can complete the entire process in parallel and automatically. High-precision three-dimensional processing of tilted images and construction of fine 3D models effectively realize the efficient production of large-scale tilted 3D models.

Features
City-level tilt data processing capability
It can smoothly complete high-precision three-dimensional processing and fine 3D model construction of more than 100,000 tilted images in a single time.

 120,000 oblique image model construction at a time

Comprehensive sensor data support
With excellent data compatibility, it can support multi-source image data shot by multiple platforms including large fixed-wing aircraft, manned helicopters, large, medium and small drones, street view vehicles, digital cameras, and mobile phones.

Highly automated workflow
From oblique image organization and engineering creation, free cyberspace three and control point adjustment, fine three-dimensional model construction, image intelligent partitioning to cluster calculation, the entire process is highly automated. Only minimal manual intervention is required in the control point prickling stage, and then you can Automatic matching predicts the position of the corresponding control point on the image with the same name, and automatically completes the control point stab work.

Stable and reliable aerial triangle operation mode
The system widely adopts universal machine vision algorithms, combined with rigorous solution based on photogrammetry, and effectively solves the problems of uneven distribution, fracture, and delamination of the connection points of the air triangle, and can refine and calculate the reliable air triangle results in a single time. The accuracy is completely better than the 1:500 air triangle accuracy requirement.

● Deeply parallel free net aerial triangulation

The system intelligently divides the task area, automatically allocates cluster processing nodes and performs in-depth parallel splitting of the air triangle processes, and performs parallel processing from feature point extraction, feature matching, free network adjustment to precise turning points, and fully improves the air triangle solutions Calculating efficiency, the obtained air-triangle effects are good and stable, and the connection points are evenly distributed.

● Forecast-based control point puncture process

1.  No point prediction: The system automatically searches for the corresponding original image based on the existing POS information and completes the work of controlling the puncture point.

2. Single point prediction: After manually piercing a single control point, the system automatically matches and predicts the position of the corresponding control point on the image with the same name and completes the point work.

● High precision adjustment with POS parameters

The POS parameters are used for the normalization of the measurement area, and the percentage of the plane error within 5cm is controlled above 70%.

In an oblique photography project in a certain area of Optics Valley, the system completed the aerial triangle processing of 20,000 images at one time, and the accuracy of the results was completely better than the 1:500 aerial triangle accuracy requirement.

A certain area of Guanggu Oblique image aerial triangulation accuracy report

Automatic reconstruction of fine 3D models
After testing on typical landforms such as flat land, hills, lakes, and urban areas, and testing in hundreds of survey areas with a scale of more than 10,000, the system modeling effect is good and stable, and the model accuracy meets the requirements of 1:500 cadastral maps.

Deeply parallel intelligent cluster computing architecture

● Intelligent task partition

For large-scale tilt data, the system flexibly adjusts and determines the task division scheme according to the position relationship between the images, the degree of coupling, and the hardware resource status of each node in the cluster, and automatically completes the intelligent partition of the large-scale tilt data, and reasonably divides many disordered images into multiple subtasks To parallel processing.

● Deeply parallel cluster operation support

The system adopts an industrial-grade parallel framework design, and cooperates with cluster hardware to support in-depth parallel splitting of the work process. At the same time, it integrates different computing resources such as CPU/GPU/FPGA of all cluster nodes in the network environment based on heterogeneous computing technology to achieve efficient distribution Parallel cluster computing.

Modeling case show

Wuhan Guanggu qicai xinghe sculpture Tilted 3D model
Guanggu International Tennis Center Tilted 3D model

 Inclined 3D model of Wuhan Outlets