📍 Time in Zone Analysis using YOLO (ONNX) + ByteTrack

This example demonstrates how to calculate dwell time (time spent in a zone) using computer vision.

It detects and tracks objects in a video and measures how long each tracked object remains inside predefined polygon zones.

This project is ideal for:

🛒 Retail analytics (checkout wait time monitoring)
🚦 Traffic management
🏭 Industrial safety zones
🛑 Restricted area monitoring
📊 Queue time analysis

🚀 What This Project Does

The system:

Detects objects using a YOLO ONNX model
Tracks objects using ByteTrack
Monitors predefined polygon zones
Calculates how long each object stays inside each zone
Displays dwell time in real-time
Saves the annotated output video

Each tracked object is labeled as:

#ID MM:SS

Example:

#5 01:24

Means object ID 5 stayed 1 minute and 24 seconds in the zone.

📥 Model Download

Pretrained models for swatahVision are available in the Model Zoo.

🔗 https://visionai4bharat.github.io/swatahVision/model_zoo/

📁 Project Setup

1️⃣ Clone the Repository

git clone https://github.com/VisionAI4Bharat/swatahVision.git
cd swatahVision/examples/time_in_zone

2️⃣ (Optional) Create Virtual Environment

python3 -m venv venv
source venv/bin/activate

3️⃣ Install Dependencies

pip install -r requirements.txt

🎨 Step 1: Draw Custom Zones

Before running dwell time analysis, you must define monitoring zones.

Use the provided script:

python scripts/draw_zones.py \
    --source_path "data/checkout.mp4" \
    --zone_configuration_path "data/config.json"

Controls While Drawing

Key	Action
`Enter`	Finish current polygon
`Escape`	Cancel current polygon
`s`	Save zones to JSON file
`q`	Quit drawing window

This generates a JSON file containing polygon coordinates.

Example config.json:

[
    [[100,100], [400,100], [400,400], [100,400]]
]

Each polygon represents one monitoring zone.

🧠 Step 2: Run Time-in-Zone Analysis

Run the detection and tracking script:

python yolov11x-1280_onnx.py \
    --zone_configuration_path "data/config.json" \
    --source_video_path "data/checkout.mp4" \
    --source_weights_path "yolov11x-1280.onnx" \
    --classes 0 \
    --confidence_threshold 0.3 \
    --iou_threshold 0.7

⚙️ Command Line Arguments

Required

--zone_configuration_path
Path to zone configuration JSON file.

--source_video_path
Path to input video file.

Optional

--source_weights_path
Path to YOLO ONNX model file.

--classes
List of class IDs to track.
Example:

--classes 0

If empty, all detected classes are tracked.

--confidence_threshold
Detection confidence threshold.
Default: 0.3

--iou_threshold
IoU threshold for Non-Max Suppression.
Default: 0.7

🧠 How It Works (Technical Overview)

Processing Pipeline:

Load YOLO ONNX model
Read video frames
Run object detection
Filter selected classes
Apply ByteTrack multi-object tracking
Check if object center lies inside polygon zone
Count frames inside zone
Convert frame count to seconds using FPS
Display timer on video
Save output video

⏱ Dwell Time Calculation

Time is calculated using video frame rate:

time_seconds = frame_count / FPS

The timer updates continuously while the object remains inside the zone.

If the object leaves the zone, timing stops.

📊 Output

The output video contains:

Drawn polygon zones
Tracked object IDs
Real-time dwell time counter
Color-coded zones
Saved output video file

🖥 Model Requirements

This project requires a YOLO ONNX model.

You can:

Export from Ultralytics YOLO
Use a pre-exported ONNX model
Use your internal trained model

Example export:

pip install ultralytics
yolo export model=yolov8x.pt format=onnx imgsz=1280

🎯 Practical Applications

Checkout wait time monitoring
Customer behavior analytics
Smart traffic signal timing
Industrial hazard zone tracking
Security restricted-area alerts
Queue performance monitoring

📌 Notes

GPU recommended for real-time performance
Zones are triggered using object center point
Supports multiple zones
Supports multiple object classes
Works best with fixed camera footage