Skip to content

JAIJANYANI/Video-Analysis-for-Surveillance

Repository files navigation

Video-Analysis-for-Surveillance

Analyzing CCTV videos to find timestamps of interesting events using deep learning

Description

Performed this analysis using image/video processing and deep learning algorithms.

The Project consist of three steps :

1.) Extracting frames from a video at regular intervals
2.) Detecting change in every two consecutive frames in a sequence
3.) Reporting timestamp for every change

Prerequisites

Software

  • Python-opencv
  • Tensorflow v 1.0 or later
  • Python 3.6.0 |Anaconda 4.3.0 (64-bit)|

Tested on Ubuntu 16.04 LTS amd64 xenial image built on 2017-09-19 8-core CPU

Installation

  • Simply clone the repository
  • Paste your video/videos in "videos" folder
  • Done

Running

  • Simply run this command from root directory.
python extract.py ./videos -o ./images --skip 75

  • NOTE : Video should be present in "videos" folder
    : Frames will be extracted in "images" folder
    : --skip = 3*FPS
    : The provided video "05_05.mp4" has 25FPS , so skip = 75

Screenshots

untitled

untitled2

untitled3

Output

start time : 0:01:39
end time : 0:02:09


start time : 0:03:45
end time : 0:03:48


start time : 0:04:27
end time : 0:04:33


start time : 0:06:57
end time : 0:07:06


start time : 0:07:39
end time : 0:08:15


start time : 0:08:18
end time : 0:08:18


start time : 0:10:45
end time : 0:10:48


start time : 0:11:36
end time : 0:11:39

Author

Jai Janyani

Licence

This project is licensed under the MIT License - see the LICENSE.md file for detail