Autonomous vehicles offer great chances to change future mobility, however these chances come along with big challenges with regard to the development and testing of the complex vehicle system. For that reason, the Formula Student Germany introduced the completely new Driverless Competition in 2017. Beside a combustion and electric vehicle, KA-RaceIng, the Formula Student Team of the Karlsruhe Institute of Technology, is now also developing a driverless race car every year.
With the help of three Basler Dart Cameras the cones that define the track should be detectetd and classified as the colour and size of the cone matter. The detection is executed with a deep neural network that takes the pictures of each camera individually and calculates a 2D position of the cone in a car coordinate system. The Network is trained from scratch with self taken and annotated Images.
Camera: Basler dart daA1600-60uc
Lidar: Ibeo Lux
Processing Unit: B-plus Car-PC
The Lidars provide cluster centroid cones by measuring distances to reflecting points. The cameras provide detections of the neural net with a classification and pose estimation. To track the cones we merge these information with a 2 dimensional projection of the cones from the camera level in the vehicle frame. The range estimation of the cones is decisively affected by the Lidar data.