Light Detector

Context: During the CIRC competition we must follow a series of red, blue, and infrared lights, while searching for crates labeled with AR tags. For full context consider reading the CIRC Traversal Mission rules or ask a lead for a verbal description.

Problem: We currently have no infrastructure for doing this task fully autonomously. Thus, you will devise a perception pipeline for detecting and publishing the locations of the lights, as well as design a algorithm which will traverse along these waypoints.

Solution: Use the ZED and perform a thresholding algorithm to locate the lights in image space. Then, using the point cloud generated by the ZED, estimate these objects in 3D space and add them to a cache. Upon each time a given location is detected with a light, increment a hitcounter. Once this counter goes beyond a threshold (will be determined with testing in sim and IRL) publish their locations to topic named "light_poses".

---

Interface (subject to change)

Node: light_detector

Subscribes: sensor_msgs/PointCloud2 (ZED point cloud)

Publishes: Vector of geometry_msgs/Vector3 (Array of Vector3)

Should Look Something Like this

Vector3[] lights

Rough Steps:

1. Use OpenCV Thresholding (over either RGB or HSV) to detect the lights in image space
2. Use OpenCV Contours to determine light image space location
3. Use the ZED's point cloud to find the corresponding 3D point to the center of the detector light
4. Create a light cache with hit counts (probably using std::unordered_map)
5. Update hit counts as new lights are detected
6. Publish custom light position message to "light_poses"

Below is a diagram which shows the perception pipeline: