1 of 1

Cameraman

C6: Jae Song, Thomas Li, Bhavya Jain

18-500 Capstone Design, Spring 2024

Electrical and Computer Engineering Department

Carnegie Mellon University

System Architecture

Product Pitch

Cameraman is a system of auto-tracking cameras and feed selection algorithm is our approach to supplement, or even replace, the current form of live car racing stream production. It hopes to decrease the human latency of feed switching as well as eliminate the danger cameramen face due to potential race accidents. Some critical requirements considered while developing this system were accuracy of tracking (car should be in the frame 95% of the time), latency of stream (camera to stream latency of less than 500ms), and quality of stream (car should be in the middle 50% of the frame 75% of the time).

​

http://www.ece.cmu.edu/~ece500/projects/S24-teamxx

System Description

System Evaluation

Conclusions & Additional Information

The system has 3 main components:

1. Video Processing Module: Detect car using color profile obtained at the start of the race and output bounding box for each frame for each camera input provided. This is provided to the selection module and the arduino
2. Feed Selection Module: Compare Bounding box sizes of frames and select the best available camera by filtering and observing past camera orderings to pre-switch.
3. System Hardware: Slot car racing track outfitted with 4 servo motor mounted cameras connected to arduino that maintain car at central position on screen

whole system

In conclusion, our system functionally runs as aspired. Although some of the use case requirements are not yet met or tested, the core functions of tracking cameras and reasonable feed selection has been successfully executed. As a team, we learned how to communicate interfaces between modules with one another and how to be open to changes in idea. In the future, this project, with more accurate detection, has potential to develop in customizability with different tracks and setups.

​

camera + motor

Camera USB to PC

Motor control

inputs

Arduino Uno

motor control

outputs

arduino -> motors

USB to PC

PC

arduino -> motors

camera + motor

power supply for track

camera POV

track

TABLE

Out of ~40 switches total, there were more switching errors with prediction but a tradeoff of being able to switch to the next camera prior to first detection. �

GRAPH�Measuring the ratio of the total frames that the car was detected to the total number of frames it was visible to a camera.

​

The camera position was varied to include straight ways, soft and sharp corners. The voltage was varied to allow car speeds around 1mps. ��Detection performs best and most consistently for straightways (~50%) whereas performance across sharp corners show a dip as speed increases.

Feed selection testing per 10 laps

Link to our website: