A custom made Labview software (using a X64 CUDA DLL for the main tracking routine) allows to track tens of molecules at hundreds of Hz and detect sub-nn displacements in real-time. The CUDA code we use is based on the one written by Shawn Tabrizi (GitHub) when working in Saleh's Lab at UCSB (based on Vincent Croquette's work). One of our Internship students (Zhenxiang Luo, now at IMEC) helped ST to run it on X64 platforms and we did additional tweaks (e.g. Flat-Field correction). Recently, Olivier Betschi (doing a 2 months internship in summer 2018, M1 INSA) interfaced our Hardware (camera, Piezo) in Python and wrote the code to call the CUDA DLL in Python. In the next few months, we will fully automize the MT device (injection of beads, recognition of (good) DNA beads, injection of relevant proteins).
Our current GPU is a MSI GeForce GTX 1060 Gaming X 3G allowing to track +50 beads (ROI: 64^2 pixels) at 400 Hz on large images (e.g. 2000 x 2000 pixels). The analysis is performed in real-time (all directions). As we use a buffered acquisition (using a Silicon Software frame grabber), there is no loss of frames.
Softwares (Python or Labview) calling the CUDA DLL and demonstrating tracking are available upon request (wilfried.grangeipcms.unistra.fr).