Feb9 Update

• Updated project block diagram

• Compiled updated document and sent to GM

• Team photo

GM mentors feedback:

Camera specs: https://www.intelrealsense.com/depth-camera-d435/

• Good: 30fps & global shutter

• Bad: max range 10m, “under ideal conditions”, according to the website.

• We won’t get any range data beyond 5m at best, and (in my opinion) < 1m during daylight hours. Keep in mind the camera to the hood is 1m, so max effective range (beyond the hood) is 4m at best.

• Pick a good stereoscopic camera instead of IR for depth perception.

Camera calibrations:

• The camera will bounce around during vehicle movement, affecting range measurement; in AVs, we use dynamic calibrations on top of calibrations found under static conditions

• https://en.wikipedia.org/wiki/Fiducial_marker An example of a fiducial marker is the checkerboard used for static camera calibrations. A car can use a hood ornament or a design painted on the hood for fiducial marker duties.

• The roof of the car is not a rigid body, thus it will affect camera vibration during vehicle movement ultimately degrading ranging performance. Mounting on the dashboard is better for rigidity, however, this degrades the range and FOV.

No guarantee of good data before efforts are spent to process and xfer the data, until 5G is available, data pipeline bottleneck is one of the major hurdles.

Safety issues with mounting cameras.

• If the vehicle crashes, those cameras may become projectiles that may hit vehicle occupants?

Backend:

• The industry standard for the backend data xfer seems to be MQTT, it’s should be compatible with ROS and I’ve seen it in quite a few IoT data collection projects.

• Yolo and OpenCV are standard for these type of projects

The software/hardware architecture is sound. Having some actual data to show proof of concept would strengthen this.