IoT, Wavelets and Machine Learning in the Smart Home

on YouTube ↗21m

" summary

Floor-mounted accelerometers stream at about 1 kHz sampling through a TCP socket into a Kafka producer and Spark Structured Streaming pipeline, where overlapping frames undergo wavelet transforms (piwavelets used) to produce scalogram images. Images flatten to ~1,120 features and are reduced via non‑negative matrix factorization to 1120→8 features for classifiers, yielding high precision and improved F1 for identity and intruder detection. Prototype is low‑cost and functional; next steps: run Kafka client on the microcontroller and move transforms into Scala.

AI-written summary · about

" chapters (11)

1. 0:00Introduction and motivation
2. 2:29Privacy and sensor trade-offs
3. 3:32Data pipeline overview
4. 4:36Hardware setup details
5. 5:41Wavelet preprocessing steps
6. 8:35Scaleograms and images
7. 10:44Dimensionality reduction
8. 13:14Results and next steps
9. 15:32Q&A: Outdoor sensors
10. 16:38Q&A: Sensor placement and edge computing
11. 18:47Q&A: Uniqueness limits

" related talks

• 33mAnalytics in the age of the Internet of Things #bbuzzLudwine Probst · Berlin Buzzwords 2015
• 56mTaming IoT Data: Making Sense of Sensors with SQL StreamingHans-Peter Grahsl · Voxxed Days Cluj-Napoca 2019
• 31mFun with SensorsYogi Rampuria · Distributed SQL Summit Asia 2022
• 40mBuilding Real-Time Applications: Cyclist Crash DetectionTomas Neubauer · Berlin Buzzwords 2023
• 32mData Science in the Internet of Things with Python and SparkRafael Schultze-Kraft · PyData Amsterdam 2017
• 20mBuilding realtime efficient queries in an IOT platformDinesh Narayanan · Data By the Bay 2016
• 49mIOT, Timeseries and prediction with Android, Cassandra and SparkAmira Lakhal · Jfokus 2016
• 01h 13mHealth Monitoring with Wireless Sensors and Machine LearningBruce M. Maggs · Algorithmic Foundations for Emerging Computing Technologies Boot Camp 2025