Einsatz der PMD-Kamera in der mobilen Robotik für die Hinderniserkennung und -vermeidung sowie für die Selbstlokalisierung

The subject of this thesis is the obstacle and driveway recognition, the obstacle avoidance and the self-localization of a mobile vehicle. In the past therefore were used 2D, 3D laser scanner and stereo vision. This thesis shows the application of the PMD camera.

As experimental platform was presented a mobile vehicle with differential drive (Tom3D) and another vehicle with Ackermann-drive (Merlin3D). Both mobile vehicles are equipped with the same system architecture, and each has an embedded PC for sensor data processing, image processing and communications, as well as a C167 microcontroller for speed control and obstacle avoidance. For the speed control and for a rough estimate of self-localization there are used wheel encoders. GPS and inclination sensors provide the self-localization. The access to such a mobile vehicle by the operator is possible as a manual control or can be realized by Wireless-LAN (WLAN). A tele-presence respectively telemetry is conceivable via mobile radio or satellite radio.

A PMD camera and a 2D camera were integrated with a defined inclination angle onto the mobile vehicle. The unambiguity interval of 7.5 m, the optical field of view and the mounting height of the PMD-camera were considered. From this perspective, the PMD camera always records obstacles and driveway together in the front area of the mobile vehicle. There must be a differentiation between obstacles and the driveway within the PMD-video image. It also applies to differentiate between so-called negative and positive obstacles. For a real-time separation of driveway, negative and positive obstacles at every PMD video image must consider the inclination angle of the PMD-camera, as well as pitch and roll angle of the mobile vehicle.

After successful extraction of the driveway and negative and positive obstacles inside the PMD-video image, it is divided into equal vertical segments. The segment (segment number) with the smallest distance will be selected.

The segment number and the minimum distance value are the linguistic input variables for the Fuzzy-Logic-Controller. After the fuzzification of these values follows the fuzzy-inference, in this are the driving manoeuvres linguistically formulated for obstacle avoidance. Finally within the defuzzification the result data are converted into concrete motor control signals. With the help of these steps navigation is realizable without environment map and without path or driveway control. The mobile vehicle thereby can be already used for uncontrolled driving or random driving for tasks of exploration.

Only with the availability of a self-localization, here specifically the relative self-localization, path driving and map generation are possible. Movements of a mobile vehicle can be observed by position changes of objects or their significant features within the continuous PMD image sequence of a PMD camera. These position changes must be determined within two sequential PMD video images. The problem is, however, to find the significant features of the first PMD video image again in the second PMD video image and assign them to each other. From these position changes the translation vector and rotation matrix respectively the pose change (position and orientation) of the mobile vehicle can be calculated.

Significant features within the participating PMD-video images can be extracted with the help of the so-called Moravec- respectively Interest-operator. Now the correspondence between two feature point sets can be done by using the Euclidean distance respectively Euclidean norm. Based on these pairings, and by using the Unit-Quaternion-Method for 3D movements or by using the HAYAI-Method for 2D movements can be determined the desired rotation matrix and thereafter the translation vector. Translation vector and rotation matrix still have errors, which can be reduced iteratively by using the ICP-Method.

Simple manoeuvres, obstacle detection, obstacle avoidance, self-localization, path driving, etc. can be combined with each other and thus create semi-autonomous functions. A hierarchical compilation will show the transition from pure remote control, via semi-autonomous and autonomous functions to useful services with the help of a mobile vehicle.