Bio-Inspired Computer Vision: Optical flow estimation using space-time separable filters
This project is one part of the module: Bio-inspired computer vision and was implemented by Bing Li under Prof. Dr. Marianne Maertens and Prof. Guillermo Gallego.
This is the code used to produce a filter part(Section 2.2 and Section 3.1) of the results published in [1] .But generating these result needs pre-knowledge included in Section 3 [2] and in Section 3.1 [3]
The repository contains the following:
-Python code: folder:src
1.optical_flow.ipynb
a.definitions of spatial and temporal filters
b. visualization of filters above in 2d or 3d view
c.implementation of equation (Eq 23) in 1[@tschechneBioInspiredOpticFlow2014] and visualization
d.using aggregation to calculate the velocity at each pixel for optical flow based on separable filters and visualization(Eq.33 1)
e. .npy files are the intermediate files2.util.py
some useful tools for loading data , the time consuming calculation, plotting tools and etc,this code is provided by Cedric Scheerlinck
-Dataset:
This is the dataset which is used in this project.For this dataset, only event.txt is used.
-Figures:
the generated figures are saved in the folder: output_figures
-Slides:
This is the folder contains the slides which explains some general ideas of this project.
Dependencies of this code
numpy,
matiplotlib,
pandas,
opencv,
scipy
References
- [1] Brosch Tobias, Tschechne Stephan, Neumann Heiko,On event-based optical flow detection
- [2] Tschechne, Stephan and Sailer, Roman and Neumann,Heiko.Bio-Inspired Optic Flow from Event-Based Neuromorphic Sensor Input
- [3]Tschechne, Stephan and Brosch, Tobias and Sailer, Roman and von Egloffstein, Nora and Abdul-Kreem, Luma Issa and Neumann, Heiko.On Event-Based Motion Detection and Integration