| The name of the base system (platform): | Artificial intelligence (AI, Artificial intelligence, AI) |
| Developers: | KB Panorama |
| Last Release Date: | 2026/01/30 |
| Technology: | GIS - Geoinformation Systems |
Content |
The main articles are:
2026: Vehicle Detection and Classification Mode on Satellite and Aerial Photos
Panorama Design Bureau has developed the Panorama Vision Automatic Recognition and Vectorization Complex version 1.9.0. This version has added a mode for detecting and classifying vehicles on satellite and aerial photographs. The developer announced this on January 30, 2026.
.jpg)
The mode is designed to work with images of ultra-high-resolution terrain (0.3-0.6 meters/pixel) and allows you to identify transport objects of the following types: cars, cars pickups, trucks buses, construction and engineering equipment (including bulldozers and excavators), sea and river vessels, aircraft.
Recognition is based on a neural network model with CenterNet architecture, optimized for highlighting point objects when analyzing dense scenes.
The model allows you to accurately predict the coordinates of the centers of transport objects, even if their size in the image is only 10-30 pixels. In addition, the applied neural network algorithms help to separate closely located objects, avoiding "gluing" cars in parking lots or ships in the port. This mode provides opportunities for automating the following tasks:
- monitoring of road situation and density of traffic flows;
- analysis of occupancy of parking lots, logistics terminals and industrial sites;
- inventory of transport and equipment at construction and infrastructure facilities;
- control of port activity and navigation;
- detection and recording of aircraft, assessment of parking lot occupancy at airfields;
Based on the results of image processing, the Panorama Vision complex forms a vector map with recognized objects. Starting with service version 1.9.0, the panvision.rsc classifier is used to create objects. Vehicles are mapped as point objects of the corresponding types. The prediction confidence value of the neural network model is written to the semantics "Prediction confidence" (code 963 in the classifier panvision.rsc). In addition to vehicles, the Panorama Vision complex allows you to identify objects of such categories as: agricultural fields, hydrographic objects (rivers, ponds, lakes and others), high forest, forest overgrowth, felling, buildings, dirt and asphalt roads. To use the new functions of the service, you need to launch the "Automatic recognition of images" task from the main menu of the Panorama GIS: Launch applications - Automatic creation of objects - Automatic recognition and vectorization complex - Automatic recognition of images. You can visualize the density of traffic flows, congestion of parking lots and logistics terminals after processing images in Panorama Vision using the Heat Map mode available in 3D Analysis Complex.
2025
"Panorama Vision" version 1.8.0 with "Search for unregistered objects" mode
Panorama Design Bureau has developed the Panorama Vision Automatic Recognition and Vectorization Complex version 1.8.0. In this version, the "Search for unregistered objects" mode has been added, designed to automatically identify objects that are not on the current digital map. The developer announced this on December 16, 2025.
.jpg)
The mode allows you to supplement and update the digital map taking into account new satellite or aerial photographs. The Panorama Vision service not only decrypts objects in images, but also compares recognized paths with reference ones preloaded to the server. As a result of the comparison, each object is assigned semantics with the corresponding display color: red - the object was found in the picture, but is not among the standards (new); orange - the object has significant geometric differences from the standard; green - the object coincides with the reference one; violet is a reference object that does not match any of the found.
The tolerance for shape deviation between the recognized object and the reference object is set in the "Search for unregistered objects" mode dialog in the Panorama GIS. By default, it is 5 meters. The deviation is calculated using the Hausdorff metric - as the maximum distance from the points of one object to the nearest points of another. The result is stored in the semantics "Maximum deviation of the shape of the object from the template" (code 31302 in the classifier service.rsc).
For the identified objects, the reference to the nearest reference object of the source map is recorded in the semantics "Reference to the map with the addition of the object number" (code 32797 in the classifier service.rsc). This mode allows you to identify objects of the following categories: agricultural fields, hydrographic objects (rivers, ponds, lakes and others), high forest, forest overgrowth, felling, buildings, dirt and asphalt roads. The processing cost depends on the area of the area of interest and the object category. Current recognition rates are listed on the product description page.
To use the new functions of the service, you need to select from the main menu in the Panorama GIS: Launch Applications - Automatic Object Creation - Automatic Recognition and Vectorization Complex and launch the "Search for Unregistered Objects" task. On the panel "Automatic recognition and vectorization complex" there is also a button for opening the dialog "Automatic image recognition," which allows you to process images of the area and data from open geoportals - even in the absence of a reference map. To get a finished vector map, the user needs to double-click on the requisition line in the Requisition List or access the context menu and select Open Layer.
The Search for Unregistered Objects mode will allow municipal and regional authorities to remotely control capital construction facilities in subordinate territories, as well as promptly register previously unregistered objects identified by the results of the analysis of images. Panorama GIS versions 15.5.2 and higher must be used to correctly operate the search functions for unregistered objects.
"Panorama Vision" version 1.7.0
On September 3, 2025, Panorama Design Bureau announced the development of the Panorama Vision Automatic Recognition and Vectorization Complex version 1.7.0.
According to the company, in the next version, the "Complex Processing" recognition mode has been added, which allows using one application to create vector maps with a set of objects from pictures: agricultural fields (arable land), water bodies (rivers, lakes and others), forests (high-barreled forest, piglets, felling), buildings and roads. As in the "City" recognition mode, the service not only finds buildings, but also determines their subtype: residential buildings (multi-storey and individual), public buildings (schools, kindergartens, hospitals and others), industrial and commercial buildings (factories, shopping centers and others), garages (garage-building cooperatives). All buildings are automatically oriented along the direction of the nearest roads. The position of the recognized contours of buildings is corrected to correspond to the foundation of structures.
To improve the accuracy of the final vector map, the Panorama Vision service automatically coordinates the boundaries of the formed area objects (forests, arable land, reservoirs, buildings) with each other. In complex processing, an additional option "Sign object addresses" is available. When choosing it on the basis of integration with the address database of GIS "Panorama" into the territory of the Russian Federation (address.gisserver.info), the service will add semantics to objects with house numbers, street names, cities and districts. Additionally, house numbers and street names will be mapped as signatures.
When you create vector map objects, you use a single classifier called map5000m.rscz. To form agricultural land objects, the Agricultural Land object with code 71324000 is used, hydrography - the Reservoir object with code 31120002. High-barreled forest is mapped as "Dense high forest" area objects with code 71111110, forest overgrowth - as "Forest overgrowth" area objects with code 71113000, logging - as "Cut forests" area objects with code 71112300. Asphalt roads are mapped as "Highway" linear objects with code 61230000, while dirt roads are mapped as "Street" linear objects with code 45200,000. Buildings are mapped as areal objects of the following types: "Residential building" (with a key a_residential_b"), "Public building" (with a key a_public_b"), "Industrial enterprise" (with a key a_industrial_b"), "Garages (GSK and others)" (with a key a_garages_b").
The Panorama Vision service allows you to process terrain images in RSW, JPG, TIFF, PNG formats, as well as data from open sources (geoportals). To ensure the quality of recognition, it is recommended to use images with a spatial resolution of up to 0.5-0.6 meters/pixel. The service is located on the Internet on the servers of Panorama KB. It is accessed from the Panorama GIS application list: Launching applications - item "Automatic image recognition," as well as using the REST API protocol methods.
"Panorama Vision" version 1.5.0 with REST API support
Panorama Design Bureau has developed the Panorama Vision automatic recognition and vectorization service version 1.5.0. The developer announced this on July 2, 2025. This version adds support for the REST API protocol for accessing the complex. Using the REST API will allow you to embed the service in web applications and desktop programs. Work with the service is carried out according to the request model.
The following methods are supported: retrieving a list of available models, retrieving a list of user-available requisitions, querying for requisition information, submitting a requisition, and others.
The complex supports the processing of terrain images in RSW, JPG, TIFF, PNG formats, as well as data from open geoportals.
Panorama Vision is able to recognize the contours of areas of fields, hydrographic objects (rivers, ponds, lakes and others), buildings and structures, roads, high-barreled forests, piglets (young tree plantations, shrubs), felling (areas with recently removed forest). The processing results are transmitted to the user as a vector map available for download using the WFS protocol. The complex is implemented in the form of services that provide remote access to data decryption. It is located on the Internet on the servers of Panorama KB. Access to the complex is carried out from the list of application tasks of the Panorama GIS or using the REST API.
"Panorama Vision" version 1.4.0
Panorama Design Bureau has developed the Panorama Vision automatic recognition and vectorization service version 1.4.0. The service is supplemented by a neural network model for segmentation of forests on multispectral (RGB) satellite images with a spatial resolution of 0.6-1.2 m/pixel.
.jpg)
The model, built on the basis of the Transformer architecture, accurately identifies three categories of objects: high forest, forest overgrowth (young tree stands, shrubs), felling (areas with recently removed forest stand). Thanks to the attention mechanism characteristic of transformers, the neural network can capture global dependencies between all parts of the image. Taking into account the context of the surrounding area allows the neural network to correctly recognize large and extended objects, such as forests, as well as more precisely classify ambiguous areas, for example, distinguish small growth from felling.
Training of the neural network model was carried out on images of the regions of the Russian Federation. The achieved segmentation accuracy on the test sample exceeds 90%. The result of image processing is a vector map of the terrain formed by the Panorama Vision service using the map5000m.rscz classifier. High forest is mapped as "Dense high forest" area objects with code 71111110, forest overgrowth - as "Forest overgrowth" area objects with code 71113000, logging - as "Felled forests" area objects with code 71112300.
Additional capabilities of the Panorama Vision service can be used in forestry for monitoring the forest fund, planning reforestation, detecting unauthorized felling and monitoring compliance with the boundaries of the allotted plots. Additionally, Panorama Vision allows you to vectorize buildings and roads, contours of agricultural fields and hydrographic objects (rivers, ponds, lakes and others). The service is located on the Internet on the servers of Panorama KB. Access to it is carried out from the list of application tasks of the Panorama GIS: "Launch applications" - item "Automatic recognition of images." Current version of GIS
"Panorama Vision" version 1.2.0 with automatic road decryption on satellite images
Panorama Design Bureau has developed the Panorama Vision automatic recognition and vectorization service version 1.2.0. The developer announced this on February 19, 2025. The program has added the ability to automatically decrypt roads on satellite images. This neural network model provides recognition of asphalt and dirt roads in images of the area with a spatial resolution of 0.6 m/pixel. Satellite images of cities of the Russian Federation were used to train the model.
Panorama_Vision.jpg)
In the decryption process, semantic segmentation of roads in the image is performed at the first stage, a raster mask is formed containing marked areas related to the road network. At the second stage, the service highlights the central road lines on the resulting mask and vectorizes road objects. During vectorization, roads are automatically "reached" to intersections with each other within the established tolerance by distance (up to 100 m). To create a vector map, use the map5000m.rscz classifier. Asphalt roads are mapped as "Highway" linear objects with code 61230000 and dirt roads as "Street" linear objects with code 45200000.
The road network map generated by the service is ready for integration into GIS. After loading the map into the Panorama GIS, the following operations become available: building a route between two points, creating an optimal route when visiting many points, building a distance graph. To perform these operations, you must start the Road Graph Editor task from the Launch Applications mode.
With the help of the Panorama Vision service, the formation of a road network map for an area of 25 square kilometers takes no more than 15 minutes. In addition to road recognition, the Panorama Vision service allows you to recognize the contours of agricultural fields, hydrographic objects (rivers, ponds, lakes and others), buildings and structures. The service is located on the Internet on the servers of Panorama KB. Access to the service is carried out from the list of application tasks of the Panorama GIS: Launching applications - item Automatic creation of objects.
2024
"Panorama Vision" version 1.1.1 with improved accuracy of decryption of contours of buildings and agricultural plots
Panorama Design Bureau has developed the Panorama Vision automatic recognition and vectorization service version 1.1.1. This version improves the accuracy of decrypting the contours of buildings and agricultural plots. The developer announced this on December 19, 2024.
.jpg)
In the complex integrated neural network is a model optimized for recognition of building contours terrain images with a spatial resolution of 0.1-0.3 m/pixel. This model improves the quality of recognition of buildings on orthophotoplanes formed by the Panorama Photo program based on the results of processing aerial photography materials. The joint use of the Panorama Photo and Panorama Vision software products allows you to automate the process of obtaining ready-made geodata as much as possible. These include orthofuels, altitude matrices, and vector maps formed to data from unmanned aerial vehicles () UAV satellite and images. In addition, the Panorama Vision complex has finalized the definition of building contours on satellite images with a spatial resolution of 0.75 m/pixel and below. For all categories of images, when forming a vector map, the algorithms for separating buildings located close to each other have been improved. When decrypting the boundaries of agricultural land, automatic "reaching" the contours of land to a rectangular shape was added. When creating a vector map, objects of agricultural land, hydrography and buildings are formed using a single classifier map5000m.rscz. To form agricultural land objects, the Arable Land object with code 71324100 is used, hydrography - the Reservoir object with code 31120002, and for buildings - the Residential Building object with code 44200000.
Panorama Vision allows you to process terrain images in RSW/JPG/TIFF/PNG formats and data from open geoportals. The software package recognizes the contours of areas of fields, hydrographic objects (rivers, ponds, lakes and others), buildings and structures. The processing results are transmitted to the user as a vector map downloaded over the WFS protocol. The complex is implemented in the form of services that provide remote access to data decryption. It is located on the Internet on the servers of Panorama KB. Access to the complex is carried out from the list of application tasks of the Panorama GIS: Launching applications - item Automatic creation of objects.
Version 1.1.0
On September 18, 2024, Panorama Design Bureau announced the development of Panorama GIS version 15.3.2 and Panorama Vision automatic recognition and vectorization service version 1.1.0. The updated version expanded the capabilities of the Panorama Vision automatic recognition and vectorization complex and improved the tools for automatic decryption of satellite images. A neural network model has been added to the program to highlight the contours of buildings in images of the area. The Panorama Vision service brings the recognized contours of buildings to a rectangular shape and adjusts their location so that they exactly correspond to the foundation of buildings. The resulting model provides recognition accuracy over 90%.
According to the company, satellite images of the cities of the Russian Federation together with vector maps containing the applied contours of the roofs and walls of buildings were used to train the model. For accurate determination of building contours, the spatial resolution of the data should be from 0.5 to 1 meter/pixel for satellite images and 0.25 - 0.01 meters/pixel for aerial photographs. For example, the recognition of building contours in satellite images of the city of Moscow with a resolution of 0.6 meters/pixel within an area of 25 square kilometers took 2 minutes 20 seconds. As a result of processing, 2895 objects were created. The use of the updated model will allow, together with data from Rossreestr, to identify unregistered buildings and structures, as well as promptly update existing information on vector maps.
The complex allows you to process images of the area in RSW/JPG/TIFF/PNG formats and data from open geoportals. Panorama Vision recognizes the contours of areas of fields, hydrographic objects (rivers, ponds, lakes and others), buildings and structures. The processing results are transmitted to the user as a vector map downloaded over the WFS protocol. The complex is implemented in the form of services that provide remote access to data decryption. It is located on the Internet on the servers of Panorama KB. Access to the complex is carried out from the list of application tasks of the Panorama GIS: Launching applications - item Automatic creation of objects. The updated version of the program and documentation are available on the Panorama KB website in the Download section.
| The site content is translated by machine translation software powered by PROMT. The machine-translated articles are not always perfect and may contain errors in vocabulary, syntax or grammar. Read original article If you find inaccuracies or errors in the results of machine translation, please write to editor@tadviser.ru. We will make every effort to correct them as soon as possible. |