Technical Skills: Python, Deep Learning, Machine Learning, Data-set collection, Data Cleaning, Neural Network training
Project Documentation: write-up file.
Libraries used:
- Tensorflow
- NumPy
- SciPy
- eventlet
- Flask
- h5py
- PIL
- python-socketio
- scikit-image
- transforms3d
- PyQt4/Pyqt5
The main objective of this project is to apply a Fully Convolutional Neural Network (FCNN) in a quad-drone. The drone identifies a person with specific features (red t-shirt in this case). Recognition is possible by training the FCNN with images from a data set. The data input comes from a drone that captures images from simulated open space.
The project runs in a unity simulation, Where it is possible to control a quad-drone that can be moved freely in the simulated open space. I set up a pattern that the drone will follow to capture images for recognition. After that, I set up spawning points where the multitude appears. The objective appears in the simulated crowd, which corresponds to the noise in the model. More information on write-up file and model training file.
This project is part of the Udacity Robotics Software Nanodegree program.