Cone Detection in Formula Student Driverless

For the 2017 Formula Student Driverless competition the Ignition Racing Team electric from Univerity of Applied Science Osnabrueck startet to build an autonomous vehicle. For this purpose we choosed the 2016 Formula Student Electric vehicle.



The Formula Student is an international design competition for students around the world. It now has three classes. Our team is competing in this competition since 2007. From the beginning in 2007 we started with an combustion engine. In 2010 we changed for an electric drive-train and in 2017 we wanted to compete with an electric-human-driven and an electric-autonomous-driven car. For the competition every team is building a new car every year. So our garage has a few used cars in it. We decided to use the 2016 electric car for the 2017 driverless project.

IR16 - Honeybadger, FSE car from 2016

Camera selection

To make the car autonomous we needed to add sensors and actuators. For the object detection we chooses to use one Basler ACE camera. To achieve our goal we needed a high framerate, a global shutter and a rgb-sensor. For easier communication we decided to use and USB3.0 camera. We ended up with the Baser ACE acA2040-55uc. To find the optimal camera position we startet a few test drives with a kart. After some tests we thought that our initial guess of a high mounted camera would be the best. In the top most position we have the highest range and the best estimate for the distance.

Camera mountet on kart for test purposes

Now we needed a good way to detect objects. In this case we focused on the cones. according to the rules left and right will have different colors. So it is helpful to have a rgb-camera. In terms of object detection we used a machine learning algorithm from the OpenCV library. For this we made our own dataset of cones and of pictures that are not containing cones. After labeling all these images we ended up with 1000 cone and 2000 non-cone pictures. With this data we startet to train a haar-classifier. Unlike neural networks this classifier is only able to detect one object at a time. Every added object for the detection will result in higher computing load. But since we only want to detect cones that is fine for us. We were able to get a good classifier out of 200 positiv samples. Only for more robustness we added another 800 positiv samples. The result of the cone detection output can be seen here:

Detected cones on a testdrive

With the results we were very happy. We could also improve the range by dropping some frames and applying the classifier in smaller areas of the image. In the end we had to find parameter for the best compromise between droppings frames and searching smaller areas.

Facing Problems

Parallel to this development we were working the actuators for the steering and the brake. Everything was looking alright until we figured out, that our battery from 2016 will not full fill the rules for the 2017 competition. We tried very hard to find a way to pass the rules. But we would need to build a new battery because of our battery management system, which was no safe enough anymore. This was not possible to us.


At this point I would like to thank Basler for the support and the opportunity to work on the topic of object detection with high quality hardware. It really made one addictive to work on such an interesting topic. For the next season our team will have a comeback in the driverless class of Formula Student. So the story isn't over and will be continued by the next seasons Team.

Bill of materials
Name Article number Link Quantity Unit Price
Basler ACE acA2040-55uc 1
Ibeo Lux 4 2
Festo pneumatic muscle 2
Maxon steering actuator 1
Total 0.00 $

Project State

Public Project


Software Licence: Project has no software
Hardware Licence: Project has no hardware

Project Tags




Does this project pique your interest?

Login or register to join or follow this project.

Back to top

Your comments, please!

Want to comment this ... Show more