 Final public deliverables |
|
|
|
|
| Specifications and Design (WP2) |
The 'User requirements report' (D2.1), can be downloaded as a public document. Both aquisition sets of the Image/Videos database are completed. The public videos can be accessed in the `Videos` section. |
|
|
|
| Mission Control System (WP3) |
|
|
|
|
| Autonomous navigation and localization (WP4) |
The goal is to perform drone localization by different sensors modalities, including global positioning (GPS), vision, and inertial sensors. The idea is to increase the reliability of the localisation in case of momentary loss of information from one of the sensors using redundant information. Also localization allows the drone to navigate more safely in autonomous modes.
CEA LIST has developed an approach based on a topological map built during a flight over the observed area. During this learning step, the UAV performs paths which are sampled and stored as a set of ordered key images, acquired by an embedded camera. The visual paths are organized as a graph providing a visual memory of the environment. Given an image of the visual memory as a target, the vehicle navigation mission is defined as a concatenation of visual path subsets, called visual route.
CEA LIST developed a framework implementing this methodology and making use of several sensors to achieve localization in real time. Data fusion is done through a particle filter using information from the camera, the gyrometers, the barometer, the magnetometers and the GPS.
Using the map constructed within the localization module, we have defined adequate control laws to compute low level commands (i.e global thrust, yaw/roll/pitch angles) for the UAV to perform autonomous navigation. |
|
|
|
| Obstacle Avoidance (WP5) |
As the first part of the obstacle avoidance module, UTUB have developed algorithms for the computation of optical flow. These Algorithms have been tested to provide acceptable performance and robustness in different environments with changing lighting conditions.
The resulting optical flow must be separated into three different classes: optic flow induced by egomotion, optic flow induced by independent moving objects and erroneous flow. UTUB has developed a robust method to detect the egomotion of the observing camera together with the set of optic flow vectors corresponding to this motion.
In combination with the egomotion we can compute the relative distance of the nearest object in any viewing direction thus allowing us to build an unstructured map of the environment. Obstacles then are avoided by evaluating the measured depths in certain viewing direction with a decision algorithm. This decision algorithm yields a new flying the direction which avoids any detected obstacle.
The second problem for the obstacle avoidance system is the detection of independent moving. The identification of independent moving objects is based on the analysis of the optic flow not induces by egomotion. Clustering of these flow vectors allows to separate optic flow induce by independent motion from real outliers in the optic flow vector field and to identify image regions corresponding to independently moving objects. |
|
|
|
|
Achievements
