Image recognition on satellite images
Satellite images are often a good base for computer vision. Usually big areas are mapped and photos are frequently updated, therefore the data is most likely up to date. Image recognition in this field is very advanced and a lot of data is freely available so the network can be trained well. Because of this, it is possible to recognize for example: roads, railways and rivers. Other objects can be detected as well, such as: cars, greenery and the status of meadows or even nature reserves. Like in many projects the challenge with deep learning is to provide correctly annotated data to train the deep learning network. Therefore it needs to know what is correct and incorrect so that the system can learn from that. Because objects often already contain GPS coordinates it is relatively easy to learn as long as the sources are usable.
Image recognition on satellite images
Satellite images are often a good base for computer vision. Usually big areas are mapped and photos are frequently updated, therefore the data is most likely up to date. Image recognition in this field is very advanced and a lot of data is freely available so the network can be trained well. Because of this, it is possible to recognize for example: roads, railways and rivers. Other objects can be detected as well, such as: cars, greenery and the status of meadows or even nature reserves. Like in many projects the challenge with deep learning is to provide correctly annotated data to train the deep learning network. Therefore it needs to know what is correct and incorrect so that the system can learn from that. Because objects often already contain GPS coordinates it is relatively easy to learn as long as the sources are usable.
Computer vision on aerial photography
Airplane pictures are often of better quality, because they are taken with a high quality camera and can also be made closer to the ground. For that reason detailed objects from pictures taken by an airplane or drone, are easier to recognize. Aerial photography are best used for fully automated mapping of a specific area. Side views can also be made with the use of oblique photos, as a result the depth as well as other details can be determined. The disadvantage of air photography is that the photos need to be taken actively, to map the whole area. This technique is best for mapping detailed objects or specific industrial zones or protected nature areas.
Detection on images made by cars
Cars that are equipped with a camera and GPS receivers can also produce Geo-referenced images. These days camera cars have high quality cameras and recently also LIDAR or infrared cameras. That’s why they are well suited for image recognition on both 2D and 3D data. The 360 degree camera images can be warped to for example top down images or can even be used directly to detect objects on the streets. With LIDAR and infrared images the 3D information can be used to gather more information about the type of material or vegetation. This will give a clear insight about the status of the road surface, street furniture, greenery and other objects.
Detection on images made by cars
Cars that are equipped with a camera and GPS receivers can also produce Geo-referenced images. These days camera cars have high quality cameras and recently also LIDAR or infrared cameras. That’s why they are well suited for image recognition on both 2D and 3D data. The 360 degree camera images can be warped to for example top down images or can even be used directly to detect objects on the streets. With LIDAR and infrared images the 3D information can be used to gather more information about the type of material or vegetation. This will give a clear insight about the status of the road surface, street furniture, greenery and other objects.