Project 08 · Computer Vision · Deep Learning

YOLO Video Detection

Real-time object detection and tracking on video streams using YOLO, capable of identifying multiple object classes with bounding boxes and confidence scores.

View on GitHub ↗

Overview

YOLO Video Object Detection implements real-time object detection on video streams using the YOLO (You Only Look Once) architecture. The system processes video frames at high speed, identifying and tracking multiple object classes simultaneously.

Each detected object is annotated with a bounding box, class label, and confidence score. The pipeline is optimized for real-time performance, making it suitable for live video feed scenarios including surveillance, traffic monitoring, and industrial inspection.

Architecture

Video Input Stream (File / Camera) ↓ Frame Extraction & Preprocessing ↓ YOLO Model Inference ↓ Non-Maximum Suppression (NMS) ↓ Bounding Box & Label Overlay ↓ Annotated Output Video / Display

Detection Demo

Real-time YOLO detection output showing multi-class object identification with bounding boxes and confidence scores on video frames.

Key Features

Real-time Inference: High-speed frame processing enabling live detection scenarios.
Multi-class Detection: Simultaneous identification and classification of multiple object types.
Bounding Box Annotation: Visual overlay with class labels and confidence percentages.
Non-Maximum Suppression: Elimination of overlapping detections for clean, accurate results.
Video Stream Support: Compatible with both pre-recorded video files and live camera feeds.
Performance Optimized: Tuned for high frame-rate processing without sacrificing accuracy.

Tech Stack

Frontend & Visuals

OpenCV Display Web Stream HUD

Backend & Pipeline

Python Stream Processing Multithreaded Pipeline

Database & Logging

SQLite Detection CSV Logs

AI & Deep Learning

YOLO Architecture PyTorch / Darknet OpenCV