G-TRISP Vehicle Detection and Tracking

Overview

This project is an intelligent traffic video analytics system developed for the G-TRISP internship task. The system performs vehicle detection, multi-object tracking, trajectory extraction, vehicle counting, speed estimation, and congestion analytics on traffic videos.

The pipeline is designed specifically for real-world Indian traffic conditions and includes custom auto-rickshaw reclassification for improved vehicle categorization.

---

Features

• Vehicle Detection using YOLOv8
• Multi-Object Tracking using ByteTrack
• Persistent Vehicle IDs
• Auto-Rickshaw Reclassification
• Vehicle Trajectory Extraction
• Vehicle Trajectory Visualization
• Vehicle Counting using Line-Crossing Analytics
• Vehicle Speed Estimation
• Congestion Analytics
• Annotated Video Generation
• CSV Trajectory Export
• Final Traffic Analytics Summary Export

---

Pipeline Architecture

Input Traffic Video
        ↓
YOLOv8 Vehicle Detection
        ↓
Vehicle Reclassification
        ↓
ByteTrack Multi-Object Tracking
        ↓
Trajectory Extraction
        ↓
Speed Estimation
        ↓
Vehicle Counting + Congestion Analytics
        ↓
Annotated Video + CSV Export + Summary Report

---

Technologies Used

• Python
• OpenCV
• YOLOv8 (Ultralytics)
• ByteTrack
• PyTorch
• MobileNetV2
• NumPy
• CSV Analytics

---

Folder Structure

gtrisp-vehicle-analytics/
│
├── input/
│
├── outputs/
│   ├── csv/
│   │   └── trajectories.csv
│   │
│   ├── detection_output.mp4
│   └── final_summary.txt
│
├── models/
│
├── src/
│   ├── classifier_inference.py
│   ├── congestion_utils.py
│   ├── counting_utils.py
│   ├── detect_track.py
│   ├── speed_utils.py
│   ├── summary_utils.py
│   ├── tracking_utils.py
│   ├── trajectory_utils.py
│   └── visualization_utils.py
│
├── run.py
├── requirements.txt
└── README.md

---

Setup

Create a virtual environment:

python -m venv .venv
source .venv/bin/activate

Install dependencies:

pip install -r requirements.txt

---

Run

Run the complete traffic analytics pipeline:

python run.py

---

Outputs

After running the pipeline, the generated outputs are stored at:

Processed Annotated Video

outputs/detection_output.mp4

Vehicle Trajectory CSV

outputs/csv/trajectories.csv

Final Traffic Analytics Summary

outputs/final_summary.txt

---

CSV Output Format

The trajectory CSV contains:

• Frame Number
• Timestamp
• Track ID
• Vehicle Class
• Detection Confidence
• Bounding Box Coordinates
• Centroid Coordinates
• Bounding Box Dimensions
• Bounding Box Area

---

Traffic Analytics Features

Vehicle Counting

Vehicles are counted using persistent ByteTrack IDs and line-crossing analytics.

Speed Estimation

Vehicle speed is estimated using trajectory displacement over time.

Congestion Analytics

Traffic congestion is estimated using:

• active tracked vehicles
• average vehicle speed
• traffic density heuristics

---

Visualization Features

The annotated output video includes:

• Persistent Tracking IDs
• Vehicle Class Labels
• Speed Estimation Overlays
• Vehicle Trajectories
• Vehicle Counting
• Congestion Analytics Dashboard