by Nick Speer

Setup

Make sure python3 is current and virtualenv is installed

Cuda and Cudnn are necessary for GPU acceleration. This current version primarily uses CPU

sudo bash setup.sh

Object Detection

This repo has an object detection module that uses the ResNet model implemented in PyTorch

object_detection_by_frame( filename )

Takes an input file such as 'image.jpg' and returns a MatPlotLib plot.

write_detection( matplotlib.pyplot, filename )

Takes a matplotlib plot and saves it to the specified filename

detect_all_frames( local input directory, local output directory )

Pass in the a local directory containing the images you want to detect and specify the output directory. This function will go through every frame in the input directory and output it as frame.jpg in the output directory. Only pass directories.

Camera

Tools to operate a camera and prepare video for object detection

Manually specify device in camera.py before using

record_video( filename , mode )

Records frames from camera device. mode is set to 'r' for recording by default. Set to 'v' to only view video feed. Press q to quit.

write_to_video( local input directory, video name )

Takes a directory of images and creates a video. The video file is large.

Video

digest_video( filename , output directory )

Takes a video and separates it into a series of frames.