Cone Detection Module

[Reschivon]

Description

This is a stub issue for the cone detection feature. Feel free to add details.

Cone position tracking is a key aspect of our stack. By knowing the relative position of each cone to the kart, we can plot the cone position in a global map and use this to generate track boundaries. We will see the same cone over multiple frames as the kart moves, and each observation will have slight individual noise. To figure out the true cone position from a smattering of observations, filters like EKF or simple averaging can be considered.

There are two main sources of raw data for sensing cones: LiDAR pointclouds and camera images. There is a Beijing FSAE team that has released their , which seems promising as a base for our first prototypes. For camera images, a promising open source YOLO model (thanks hayagreev). The final goal will likely be fusing measurements from both sensors.

Sub-tasks

• As an early prototype, simply getting a stock algorithm to run on the stack is sufficient
• Then we can think about integration
• Feel free to add steps as they come up.

Resources

• Check links above...