Real-time signal processing
Low-latency audio and inertial-sensor pipelines: FFTs, spectral analysis and multi-pitch estimation running at interactive frame rates on mobile hardware.
Timo Rantalainen, PhD · Docent
Numerical analysis that runs in real time, on the device in your hand.
I build measurement tools for sport and health technology: signal processing, computer vision and applied machine learning, taken from a Python prototype to Android and iOS apps that run in real time on the phone.
15+ yearsof biomechanics and sport science research
168peer-reviewed articles, 4,500+ citations
Python · C++ · JavaAndroid NDK, OpenCV, Vulkan compute, ONNX
Open sourceBoneJ, ImageJ plugins, Android apps
What I do
Computer vision and estimation
Camera calibration, sub-pixel image alignment, rolling-shutter modelling and nonlinear least-squares fitting of physical models, with honest uncertainty estimates.
Applied machine learning
Labelling custom data sets, fine-tuning detectors such as YOLOX, and deploying the trained models to phones through ONNX and OpenCV DNN.
Research-grade validation
Methods that are written up, cross-validated between devices and reproducible. Measurement tools that stand up in peer review as well as in the field.
Featured work
Flight Plate Monitor & Control
Phone-based disc golf launch analysis: speed, spin, launch, nose, hyzer and wobble from 120 fps video, using a retrained YOLOX detector and a 17-parameter rolling-shutter fit.
Read moreAudio FFT Visualise
Real-time spectrum and multi-pitch display for string and pipe instruments, with low-latency native audio and GPU (Vulkan) compute for the FFT and pitch salience.
Read moreOpen-source imaging tools
A pQCT bone analysis plugin, partly integrated into BoneJ, and ImageJ tools for segmenting MRI slices.
Read moreGet in touch
I am available for consulting on algorithm development, sensor and video analysis, and sport technology product development.