CargoTrackAI

CargoTrackAI is a computer vision project that demonstrates how deep learning--based object detection can be integrated into a structured object counting system using rule-based motion analysis.

This project covers the complete pipeline:

• Extracting frames from video
• Labeling images using annotation tools
• Training a custom YOLO model via command line
• Running inference on video using a Python script
• Counting detected objects using motion logic

The implementation is fully done in Python.

---

📁 Project Structure

CargoTrackAI/
│
├── dataset/            # Labeled dataset (images + labels)
├── original data/      # Raw source videos
├── train/              # Training configuration files
├── runs/detect/        # YOLO training outputs
├── objects/            # Extracted frames or processed images
│
├── extract_frames.py   # Script to extract images from video
├── test_model.py       # Script to run inference + counting
├── req.txt             # Python dependencies
├── README.md
├── LICENSE
└── .gitignore

---

🚀 Workflow Overview

1️⃣ Step 1 -- Extract Frames from Video

Raw videos were placed inside:

original data/

Frames were extracted using:

python extract_frames.py

This script: - Reads video using OpenCV - Extracts frames at a fixed interval - Saves images into the dataset folder

---

2️⃣ Step 2 -- Image Labeling

After extracting frames, images were labeled using an annotation tool.

Tools You Can Use

• Roboflow (Web-based)
• LabelImg (Desktop)
• CVAT
• Any YOLO-compatible labeling tool

Label Format

Images were labeled in YOLO format:

class_id x_center y_center width height

All values are normalized between 0 and 1.

The dataset folder follows this structure:

dataset/
 ├── images/
 │    ├── train/
 │    └── val/
 └── labels/
      ├── train/
      └── val/

---

3️⃣ Step 3 -- Model Training (Command Line)

Training was done using Ultralytics YOLO via command line.

Example training command:

yolo detect train   model=yolov8n.pt   data=train/data.yaml   epochs=50   imgsz=640

After training completes, model weights are saved in:

runs/detect/train7/weights/best.pt

---

4️⃣ Step 4 -- Testing the Model (Manual Video Input)

Inference is performed using:

python test_model.py

This script:

• Loads the trained model
• Accepts manual video input
• Runs object detection
• Tracks object movement
• Applies rule-based counting logic
• Displays processed video with:
  • Bounding boxes
  • Counting lines
  • Live object count

---

🧠 Counting Logic

The system uses two horizontal reference lines:

• Lower line → Entry zone
• Upper line → Exit zone

Counting rule:

1. Object crosses lower line → marked as inside
2. Object crosses upper line → count incremented
3. Each object is counted only once

This prevents duplicate counting.

python count_conveyor.py

---

🛠 Installation

1. Clone Repository

git clone https://github.com/Darshan-U-P/CargoTrackAI.git
cd CargoTrackAI

2. Install Dependencies

pip install -r req.txt

---

▶️ Run Inference

python test_model.py

Make sure your trained model path inside test_model.py is correct.

---

📦 Requirements

Major dependencies:

• Python 3.9+
• ultralytics
• opencv-python
• torch
• numpy

See req.txt for full list.

---

🎯 Purpose

CargoTrackAI demonstrates:

• End-to-end custom object detection training
• Practical dataset preparation
• Real-time video inference
• Rule-based object counting logic
• Structured computer vision pipeline implementation

This project is designed for learning, experimentation, and demonstration of applied AI in monitoring systems.

---

graph

---

ℹ️ confusion matrix

---

result

1.mp4

---

📜 License

This project is licensed under the MIT License.

---