04 / 14 — Computer Vision Suite
High-Speed Object Detection YOLOv8
General-purpose object detection pipeline with DeepSORT multi-target tracking for high-throughput video.
high-speed-object-detection-yolov8.build
Overview
What it does
General-purpose object detection pipeline with DeepSORT multi-target tracking for high-throughput video.
The Problem
Why it needed building
Detecting objects frame-by-frame isn't enough for video analytics — without consistent tracking, the same object gets treated as a new one every time it's briefly occluded.
The Approach
How it was engineered
Paired a YOLOv8 detector with DeepSORT tracking on a CUDA-accelerated pipeline, exposed through FastAPI for batch video ingestion.
The Outcome
Where it landed
The pipeline processes at 120 FPS in batch mode with a 94.2% Multi-Object Tracking Accuracy score and under 1% identity-switch rate.