Detection, Clasification and State Estimation of traffic related objects
Goals
Cognitive Automobiles will be capable to perceive their enironment an to interpret the behaviour of other traffic participants to automatically generate suitable behaviour and to cooperate with others in their perception process and decision making. The subproject A2 aims to deliver a comprehensive internal model-based description of the actual traffic-related scene. The vision based approach combines shape and dynamic models and emphasis on robustness and realtime capability of image vision algorithms. Multifocal camera platforms for active vision also enclose farfield objects on highways and rural roads. Therefore camera gaze control is another important goal.
Methods
- Edge-detection based algorithms for feature extraction, extending to color- and texture-based algorithms in future research work
- Incremental object state estimation based on a 4D-approach - an internal model of shape and dynamics is hold and updated with new image information.
- Object classification by interpreting shape and dynamic model information.
- Estimation-based, multifocal, saccadical vision with gaze-controlled cameras and inertial stabilized tele cameras to improve the image information content and algorithms.
Role in scope of project
The subproject A2 can be seen as the second vision based percepion module. Complementary to the Subproject A1 the focus is more about a multifocal farfield detection of traffic-related objectes. The output will be available to all other subprojects through a realtime database (C3).