/Human-Tracker

Identify Humans from side view CCTV Footage using a self designed Neural Network

Primary LanguageOpenEdge ABLMIT LicenseMIT

Pedestrian-detection

Pedestrian detection on Raspberry Pi using Neural Nets

Features

  • Trainable for other custom objects instead of Pedestrians

  • Real-time detection

    Dataset features

    • TLD tracker(from OpenCV) can be used to extract cutom objects from videos, (requires C++)
    • Images can also be augmented (flipped/brightness/contrast) to increase dataset size

Working

  • Train the system on the PC
  • Copy the saved weights onto the Pi
  • Two threads execute during runtime
    • One reads images from camera
    • Second performs detection
  • To speed up computation, moving objects are exctracted using difference of frames
  • Extracted objects are passed to neural net for classification
  • If it is the required object, it is marked and displayed

Dependencies

  • Python
  • Numpy
  • OpenCV

Sample Result

Image