Machine Vision is a valuable and proven instrument for measurement and control in many fields of automation technology and robotics. Image processing systems may also offer interesting solutions for flight systems:
Quick and convenient explorations of image processing methods can be performed using a powerful framework DIP. The user profits from easy access to standard tasks of image processing like image acquisition from various sources, monitoring and replay, interfaces to image processing libraries etc.. Experimental algorithms or new filters can easily be integrated. A graphical user interface (SPICE) helps to combine and tune the elements of the image processing chain. After the development process the algorithms can be directly used on a separate real-time image processing computer. Via remote control important parameters of the algorithms can be changed during runtime.
Feature Extraction and Tracking
Feature extraction is a basic task of image (pre-) processing to "understand" image contents and to reduce the amount of data for realtime use. For this purpose mature algorithms have been implemented and optimised. Also different methods for tracking and motion analysis have been surveyed and tested for their suitability for flight systems. Applications for terrain-following, egomotion measurement and navigation improvement have been demonstrated.
Color Processing
For classical image processing problems often just gray values are considered. Some Tasks like horizon detection or special search missions can be (better) met using colour information. For this reason we also work on the interpretation of colour or bayer images.
Stereo Vision
Via stereo processing of image pairs cameras can be turned into real 3-D sensors, providing depth or distance information for each pixel. These ability can be used for scene recognition, mapping applications or collision avoidance. DIP is able to process stereo data and to generate depth images in realtime. Ongoing research concentrates on making stereo measurements more reliable and less susceptible to environmental disturbances.
A major drawback of image processing sensing is a distinct illumination dependency. The range of application can widely be expanded when infrared images are considered too. Possible applications are nigh time operations, rescue missions with smoke, foggy or cloudy environments, and detection of other aircrafts.
Simulation environment
Complex image processing chains from cameras, computing hardware, drivers, filters, algorithms and their interfaces demand elaborate adaptations and tests. For this need a comprehensive simulation environment has been built up. It consists of a visualization computer to generate the scenery. This scenery is displayed on TFT monitors and captured with the original camera and image processing computer. There also exists the possibility of closed loop simulations, which facilitates sophisticated test scenarios.
By tools for image based motion analysis video sequences can be used for subsequent evaluation. These tools have already been applied for various measurements where conventional sensors are not applicable, for example under harsh conditions or where presence of discrete transducers is undesirable.
Mosaiking The mosaiking Process is used to combine multiple detail images to one high-resolution overall picture. Obviously this is a convenient method for aerial explorations. Starting with mature algorithms for the seamless combination of single images we added a method which enables an optimized image acquisition. Using this method it is possible to guarantee sufficient overlap of the images and minimal amount of data at the same time (patent pending). This method does not use position and attitude data but exploits the contents of the image. Hence no additional sensors or accurate preflight planning are needed. Mosaiking Demo Video
The mosaiking Process is used to combine multiple detail images to one high-resolution overall picture. Obviously this is a convenient method for aerial explorations. Starting with mature algorithms for the seamless combination of single images we added a method which enables an optimized image acquisition. Using this method it is possible to guarantee sufficient overlap of the images and minimal amount of data at the same time (patent pending). This method does not use position and attitude data but exploits the contents of the image. Hence no additional sensors or accurate preflight planning are needed. Mosaiking Demo Video
For development and testing following facilities can be provided: