- Home
- Skills
- APIs & Backend
- YOLO SORT Tracker — Agensi Skill Pack
Works with the AI Agents you already use
YOLO SORT Tracker — Agensi Skill Pack
Implement and debug end-to-end computer vision apps with YOLOv8 detection, SORT tracking, and FastAPI streaming.
$4.50
YOLO SORT Tracker — Agensi Skill Pack
Example session with this skill installed
My SORT tracker is losing IDs as soon as someone walks behind a thin pillar for two frames. How can I adjust the configuration in my FastAPI tracker setup to fix this?
- Read your context and instructions
- Compiled the yolo sort tracker
The issue is likely that the track is being aged out too quickly or the motion prediction isn't overlapping enough with the detection when the person reappears. Since the occlusion is short (2 frames), we can adjust the SORT_MAX_AGE and SORT_IOU_THRESHOLD.
SORT Tracking Calibration
| Parameter | Action | Reasoning |
|---|---|---|
SORT_MAX_AGE | Increase to 5 or 10 | Allows the Kalman filter to maintain the track state for more frames without a detection. |
SORT_IOU_THRESHOLD | Decrease to 0.1 or 0.2 | Makes the association more lenient when the predicted box and detection have low overlap. |
SORT_MIN_HITS | Keep at 3 | Ensures tracks are stable before they are confirmed to the UI. |
To implement this in your tracking logic, update the initialization of your tracker:
# tracker_module.py
from sort import Sort
# Increase max_age to handle short occlusions
# Lower iou_threshold if detections are jittery upon reappearance
tracker = Sort(
max_age=10,
min_hits=3,
iou_threshold=0.15
)
Next steps
- Update the
max_ageparameter in your tracker initialization and restart the FastAPI server. - Monitor the WebSocket statistics to see if the ID remains consistent after the 2-frame occlusion.
- Verify that increasing
max_agedoesn't lead to "ghost tracks" persisting too long in open areas.
Connects securely to your tools. The creator never sees your data.
About this skill
Building real-time computer vision systems involves a complex dance between hardware, inference loops, and frontend latency. This skill pack provides the specialized engineering support needed to build and debug end-to-end applications using YOLOv8, SORT tracking, and FastAPI.
What it does
- YOLO Pipeline Optimization provides guidance on model selection, confidence filtering, and precision-recall trade-offs for YOLOv8 workflows.
- SORT Tracking Integration implements multi-object tracking using Kalman filters and Hungarian matching for stable identity persistence.
- Backend Architecture designs thread-safe frame processing loops and FastAPI streaming endpoints for MJPEG and WebSockets.
- Performance Diagnosis identifies bottlenecks in FPS, latency, and resource utilization across CPU and GPU environments.
- Frontend Connectivity debugs React and Vite integrations to ensure smooth live video rendering and statistics display.
How it works
- Analyze Architecture by inspecting your existing Python project structure and frontend-backend data flow.
- Configure Detection by tuning YOLO thresholds and class IDs to match specific scene requirements.
- Calibrate Tracking by adjusting
SORT_MAX_AGE,MIN_HITS, andIOU_THRESHOLDto prevent identity switches. - Debug Streaming by tracing frame capture through to MJPEG browser rendering and WebSocket updates.
Frameworks & tools
This skill supports Python 3, OpenCV, Ultralytics YOLOv8, FastAPI, NumPy, SciPy, and React/Vite with Tailwind CSS.
Why this beats prompting it yourself
Generic AI often confuses vanilla SORT with Deep SORT or fails to handle thread-safe state in FastAPI loops. This skill maintains strict separation between motion-based tracking and appearance embeddings while ensuring correct NumPy-based detection formatting.
Use cases
- Building a live dashboard for real-time people counting via webcam.
- Debugging ID fragmentation in high-density multi-object tracking scenarios.
- Transitioning a local CV script into a production FastAPI service with a React UI.
- Optimizing YOLOv8 inference speed for edge devices with constrained hardware.
Known limitations
Does not support appearance-based re-identification (Deep SORT) or persistent trajectory databases. Requires hardware context to provide accurate performance estimates.
Requires user-provided hardware specs for FPS advice. No support for Deep SORT appearance embeddings. Does not include track persistence or trajectory export logic.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed today
Requires