PROJECT 02 · PERCEPTION

ROS 2 Vision Perception Pipeline

Real-time object detection, persistent multi-object tracking, and instance segmentation using pretrained YOLO with GPU acceleration.

31.95Average observed FPS
18.26 msAverage processing latency
24.87 msP95 processing latency
8.70 msAverage YOLO inference latency
ROS 2 HumbleYOLOTrackingInstance SegmentationCUDA
Demo

Real-time perception

The recorded demo will be embedded from YouTube here so reviewers can inspect the system immediately.

YouTube embed slot — Vision demo
Measured result

30 Hz real-time operation

2,136 steady-state frames were analyzed. P95 processing latency remained below the nominal 33.3 ms frame period of a 30 Hz input stream.

Evidence available

  • CSV performance log
  • Jupyter benchmark analysis
  • FPS / latency / object-count plots
  • Recorded ROS 2 demo
  • Source code and launch configuration