Vision is a powerful sense, plenty of information that can be extracted to be used for on-line and off-line decisions.
Unmanned Aerial Vehicles (UAV) are ready to step in the civil applications arena, including a wide range of applications, such as security, environmental, agricultural, imaginary, mapping, monitoring and inspection.
We demonstrate how Computer Vision can successfully be exploited to significantly enlarge the possibilities of using UAV in civil applications, as image stabilization, image mosaicing, flight control, visual guidance, 3D reconstruction and aerial image enhancement.
Image Enhancement
The images taken from an aircraft can be assembled together in order to obtain a huge image composed of pieces of the original images. For this propose every single image have to be referenced to the same coordinate system, that implies an accurate matching and calculation of the point of view of every single image. The accumulative error can be solved when the aircraft returns to take images of previous visited places, then the whole calculation is updated so that the new images of the revisited placed match the previous ones.
Aerial images taken from aircrafts present often big jumps caused by the aircraft vibrations and quick movements, which often make them not suitable neither for a relaxing human inspection nor for automatic visual analysis. This problem is solved by preprocessing the images, so that every new frame is matched with previous one and warped in order to obtain a stable image sequence.
Scene Inspection
By combining different computer vision techniques, such as motion estimation and image alignment, applied to information extracted by cameras, we manage to track objects in the scene, overcoming the constraints of real-time operations. We test the performances of these approaches in Unmanned Vehicles (aerial/ground).
By means of employing computer vision techniques, it is possible to perform visual inspections of natural environments, areas with the purpose of identifying and recognizing. This is useful in a wide array of activities of civilian nature.
We study the problem of the interpretation of a scene from a collection of basic features, such as lines and points, or more complex ones, such as SIFT or SURF features that are automatically extracted form single images and are tracked throughout an image sequence. These features are the basis of the visual SLAM algorithm for the localization and mapping from aerial images.
Visual Control
It is oriented to used vision-based control strategies to the object following problem, combining the image processing and control techniques in such a way that the visual information is used within the control loop. The proposed techniques will consider the use of the UAV as the object to control, or/with the onboard pan and tilt camera platform. The Inspection and Recognition techniques are used for the object detection.
By combining different computer vision techniques, such as motion estimation and image alignment, applied to information extracted by cameras, we manage to track objects in the scene, overcoming the constraints of real-time operations. We test the performances of these approaches in Unmanned Vehicles (aerial/ground).
