Usage of gait information to establish identity of a person in traffic cameras. Re-identification of persons across multiple cameras using their gait information.
#Testing out our work:
- Clone this repository on your desktop
- Download the MoBo database from the official CMU website .
- Make a folder called Mobo and extract the archive called moboBgSub.tgz in this directory
- Open MATLAB or Octave and browse to the root of this repository.
- Run the script centroid_extraction.m to get clustered centroids of all images in fast_walk pose.
- Form the feature vectors by taking the abcissa/ordinates of each of the 10 centroids.
- Use the net.m script to train the Nonlinear autoregrssive neural network on the set of observations.
Note: A seperate network to train each of the vectors is more advisable rather than taking a matrix of all centroids and its trend in the timestep.
TODO:
- Image clustering to get positions of head, torso and 10 other key features.
- Organizing the clusters and optimizing it so that the clusters represnt constant classes in all images
- Time series analysis on the cluster centroids to observe lag in data
- Seperate trend, seasonality and stochasticity of the data
- NAR based Neural network to train on the time series
- Predict based on the trained neural network
- Test results plotted in performance curves
- Develop dynamic kernel using the time series estimates.
- Calculate simmalirty score using p-values and check if hypothesis is true.