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

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