jin-qin/TrackPedesANN

Using deep neural network to detect and track pedestrian in real-time

TrackPedesANN

Using artificial neural network to detect and track pedestrian in real-time

Main idea

• Using Deep Neural Network to detect pedestrian.
• Using Convolutional Neural Network to track the pedestrian, make sure the model will not lose the detected pedestrian.
• Using Recurrent Nerual Network to predit pedestrian's position in next frame?

Language and platform

• Python
• x64 system

Schedule

• Have read all references?
• Have determined the final architecture?
• Starting coding?
• Release a demo?
• Release the final version?

Architecture

• Deep learning framework
  • Tensorflow(more flexible, slower)
  • Caffe(easier to use, faster)
• Initialization Method
  • Gaussian
  • Orthogonal initialization
  • LUSV initialization
  • Xavier initialization
  • Kaiming He initialization
• Preprocessing Method
  • PCA or ZCA Whitening
  • Substract mean images
  • Normalization

Wait for determining

• Recurrent Neural Network?
• Residual Neural Network? (The newest neural network architecture, much faster and deeper)
• Do we need an onlie tool for communication? Such as Gitter, Slack or others.