Interactive Wand

A personal passion project recreating the magic of spellcasting through computer vision, machine learning, and themed show control — all powered by a Raspberry Pi 5 and written entirely in Python.

Webpage Link: https://andrewcongdon14.wixsite.com/andrew-congdon/interactive-wand

---

Project Summary

This wand system detects spellcasting gestures in real-time using OpenCV and an infrared-lit wand. It recognizes and responds to two specific spells:

• "Alohamora" — opens the magical box with warm purple fire
• "Colloportus" — closes it with a cool burst of blue flame

The system features:

• Real-time IR blob tracking and wand path tracing
• Spell recognition using a trained SVM classifier
• Servo-based box movement
• Custom LED animations tied to spell type
• Themed sound effects with seamless background music
• Filtering to prevent false or accidental spell detection

All code runs on-device using multithreaded Python and a Pi Camera.

---

Technologies Used

• OpenCV for video input and motion tracking
• scikit-learn SVM with GridSearchCV for spell classification
• Pi5Neo to control RGB LED strip over SPI
• pygame for real-time sound effects and music
• pigpio and gpiozero for hardware PWM and servo control
• Custom wand trace dataset of 400+ samples, labeled and trained manually
• Threading to keep vision, servo, LED, and audio systems responsive

---

Spellcasting Flow

---

File Overview

HarryPotterWandcv.py

↳ Main runtime script: blob detection, trace drawing, spell prediction, and show control.

HarryPotterWandsklearn.py

↳ Used to run the pre-trained SVM classifier concurrently.

new_custom_classifier.pkl

↳ Pre-trained model for classifying spells based on trace shape.

lastframe.jpg

↳ Latest wand trace visualization, saved for debugging or training.

Sounds/

↳ Sound effects and background music used in spellcasting.

DatasetCreation/

↳ Python for drawing custom training data, converting that training data into the correct format, training the SVM classifier to produce the .pkl file

---

ML & Classification

I created a custom dataset by collecting over 400 wand path traces drawn in-air. These were:

• Centered and normalized
• Smoothed and resampled
• Converted to vector features

I used GridSearchCV to tune a Support Vector Machine (SVM) classifier that could distinguish between gestures with over 99% accuracy.

The classifier runs on-device in real time with minimal latency.

---

Show Control Highlights

• Servo Logic – Smooth actuation of box lid using hardware PWM and pigpio
• LED FX – Custom “fire” animations with randomized color flickers using Pi5Neo
• Audio Layers – Spell SFX mixed over looping background music via pygame
• Gesture Filtering – Start and stop conditions prevent noisy traces from triggering spells

---

🎥 Demo Video

Click the image to watch the full demo.

---

Final Thoughts

This was one of the most technically rewarding projects I've created — combining embedded hardware, computer vision, machine learning, and interactive storytelling. It’s a small glimpse into how software and show control can bring magic to life.