The cs293b-project from sctsui

cs293b-project's Introduction

Edible Campus Human Recognition Project

We use TensorFlow to classify videos from UCSB's Edible Campus program with the aim of determining whether humans are present at the garden or not. We envision that our solution might be implemented to guard against theft or vandalism.

Scripts Usage

Monitoring Script

Overview and usage:
- Bash script for use on IoT device.
- Extra requirements: sudo apt-get inotify-tools.
- Usage: monitor.sh [DIR] to monitor a directory.
- Executes Python classification script when a file with a .mkv extension is placed in DIR.
- When a human is detected, an email is sent to the account specified in detect_human_monitor.py.
Download the pre-trained model:
- wget http://download.tensorflow.org/models/object_detection/faster_rcnn_inception_v2_coco_2018_01_28.tar.gz
- tar -xvzf faster_rcnn_inception_v2_coco_2018_01_28.tar.gz
- Also available here (the important component is the frozen inference graph).
Modify variables in detect_human_monitor.py:
- threshold - percent of human within a frame.
- humanCountThreshold - number of frames in the video that should contain humans.
- videos_path - directory path which contains the videos.
- model_path - should point to the downloaded and extracted model from above step.

Video Transfer Script

Overview and usage:
- Simulates the arrival of videos on the IoT device.
- A new video arrives every five seconds.
- Usage: add_videos.sh [DIR] to add videos to DIR.
- DIR should be the same directory specified in the monitoring script's command line argument.
Modify variables in add_videos.sh:
- contains_humans_dir - directory path for videos manually classified as containing humans.
- human_less_dir - directory path for videos manually classified as not containing humans.