Autonomous racing vehicles
At FaSTTUBe, the Formula Student Team of the TU Berlin, nearly 80 students build racing cars every year to take part in international Formula Student design competitions. Since 2019, a system that enables the car to navigate racetracks completely autonomously is being developed and installed. Because the tracks are characterized by tight turns and chicanes, low latency and high image quality are very important for visual navigation. Last season brought a breakthrough for the FaSTTUBe team - supported by Allied Vision Alvium CSI-2 cameras.
The challenge
Low latency image transfer
In autonomous mode, the car locates the drivable route by means of cones that are set up at regular intervals along the edge of the route. To localize the cones as precisely as possible, it is very important to capture images with minimum distortion and to send them to the graphics card of the connected NVIDIA Jetson AGX Xavier with low latency. There, the images are efficiently preprocessed, and the cones are extracted and located in space using neural networks. However, an initial prototype based on USB cameras proved too slow for the highly dynamic environment of a race car.
The solution
High quality images
As an essential part of the autonomous pipeline, the FaSTTUBe Team relied on two Alvium 1800 C-319c cameras, which are particularly characterized by a global shutter sensor and a MIPI CSI-2 interface. This shutter technology prevents a rolling shutter effect at high speeds, reducing distortion. The interface allows the images to load directly into the graphics card through the V4L2 drivers provided by Allied Vision.
Furthermore, the Alvium cameras were chosen because of their very compact form factor, which proved to be an advantage in the design of the enclosed waterproof sensor box.
The benefits
Higher precision
Allied Vision's cameras have enabled a new computer vision concept that has allowed the FaSTTUBe team to not only increase the frequency of cone detection by 150%, but simultaneously increase the detection range by 400% due to the high quality of the image sensor. These increases have allowed them to make great progress in "Driverless" during the season.
The camera: Alvium 1800 C-319
Small, high-performance industrial CSI-2 camera
Alvium 1800 C-319 with Sony IMX265 runs 54 frames per second at 3.2 MP resolution. It offers high bandwidth by using the powerful MIPI CSI-2 interface with up to 4 lanes. On-board ISP functionalities enable advanced image corrections and free up the embedded board for other processing tasks. Thanks to its compact form factor and its intelligent power management it can be integrated into low-power or even battery-powered systems.
Highlights at a glance
- Easy integration into the design of the enclosing waterproof sensor box
- Distortion free images thanks to global shutter
- Increased frequency of cone detection by 150%
- Increase in detection range by 400%