The StoryTeller: Scalable Building- and AP-independent Deep Learning-based Floor Prediction
summary
System converts device WiFi scans into WiFi-scan images, normalizes them into a virtual building footprint, and feeds those images to a VGG19-based CNN for floor classification. Implemented and tested on three buildings with data from 20+ users, models trained in one building transfer to others and achieve at least 98.3% within one floor accuracy, outperforming prior methods by about 38% on the error CDF, showing a reusable, low-overhead floor ID solution that works in practice.
AI-written summary · about
related talks
01h 07mDeep Learning with TensorFlow
17mGeoAI: a feature extraction tool set using Tensor Flow and Keras...
35mAnomaly Detection in Smart Buildings using Federated Learning
06mTeaching RF to Sense without RF Training Measurements
57mTensorflow, deep learning and modern convolutional neural nets, without a PhD
39mInsights About Places with Deep Learning Computer Vision
39mImage Recognition DL WORKSHOP
33mWhat and where? - Machine learning for geospatial image analysis