Automated driving is one of the key technologies for the future transportation sector. Therefore, the complete development chain from environment detection and motion planning up to intelligent control methods has to be considered. A robust detection of path boundaries enables online motion planning of the ROboMObil. More specifically, the ROboMObil should be able to detect cones in its environment and localize them precisely. For perception the ROboMObil will be equipped with a stereo camera and a LiDAR sensor. Also the sensor data of the LiDAR and camera can be fused in order to increase the accuracy of the cone localization.
The cone detection is conducted on the 2D colour images provided by the camera. In order to enable a correct detection of cones, the images first need to be pre-processed. Afterwards, regions of interest (ROI) based on colour are defined in order to only examine areas of the image which might contain a cone and are worth further processing. In the final step, a template matching method is applied in the ROIs. The main steps of the cone detection approach can be summarized as:
- Colour model conversion
- Colour filtering
- Noise filtering
- Defining regions of interest
- Template matching in the ROI
An HSV image of two cones sent into the colour filter and the resulting binary image of two obtained after colour filtering: