MCMF-Trajectories.jl

Compute trajectories using a minimum-cost maximum-flow approach

Example

Using a pre-tracked set of Beetle trajectories the algorithm visually performs as follows

The Pre-tracked set

Randomly permuted trajectories fed into the algorithm

Output of the tracking algorithm

• ~ 5 minutes for the whole process
• ~ 70% of matrix elements correct

Original Paper for the Trajectory Framework Implemented

Zhang, L., Li, Y. and Nevatia, R., 2008, June. Global data association for multi-object tracking using network flows. In Computer Vision and Pattern Recognition, 2008. CVPR 2008. IEEE Conference on (pp. 1-8). IEEE

TODO

• Organise a full into module
• Proper error/accuracy measurement
• Add parameter tuning?, neural network?

Base Minimum-Cost Maximum Flow Algorithm

A.V. Goldberg, "An Efficient Implementation of a Scaling Minimum-Cost Flow Algorithm," J. Algorithms, vol. 22, pp. 1-29, 1997

This code uses a ported C++ Implementation of the Original C MCMF Algorithm by Cristinel Ababei January 2009, Fargo NDcristinel.ababei@ndsu.edu

CS2-CPP

Files edited from CS2-CPP to fit with Julia mcmf.cpp mcmf.h

C++ code ported using CxxWrap.jl