Object Detection: Pixels to Perception

This session covers modern Object Detection techniques with hands-on implementations using YOLO, SSD, and a real-world traffic monitoring system.

📂 Contents

1️⃣ Image_Detection_using_YOLO_algorithm.ipynb – YOLO

• Loaded pretrained YOLOv8 model (Ultralytics).
• Performed detection on online and custom images.
• Displayed bounding boxes, class names, and confidence scores.
• Demonstrated one-stage, real-time object detection.
• Printed structured prediction outputs.

2️⃣ Pedestrian_Detection_Using_SSD.ipynb – SSD

• Used SSD300 (VGG16 backbone) with transfer learning.
• Built custom PyTorch dataset (PennFudanPed).
• Generated bounding boxes from segmentation masks.
• Trained and optimized the model for pedestrian detection.
• Visualized predicted bounding boxes on test images.

3️⃣ Smart_Traffic_Congestion_Estimation_Using_Object_Detection.ipynb – Traffic System

• Applied YOLOv8 with object tracking on traffic video.
• Counted vehicles crossing a virtual line.
• Classified traffic density (Low / Medium / High).
• Generated processed video output.
• Exported vehicle log as CSV file.

📊 Slides – Object Detection Overview

• Faster R-CNN architecture (Two-stage detector).
• YOLO: One-stage real-time detection.
• SSD: Single-shot multi-scale detection.
• Anchors, bounding boxes, and class prediction concepts.
• Model comparison and key takeaways.