DOMAIN SPECIFIC ANOMALY DETECTION IN VIDEO SURVEILLANCE USING TRANSFER LEARNING

Implementation of Real-world Anomaly Detection in Surveillance Videos paper from CVPR 2018. Original implementation from authors is here.

Download C3D sports-1m weights from here and save them to 'trained_models' folder as 'c3d_sports1m.h5'.

Run demo.py to run the code on the demo video in 'input' folder. Visualization of the predictions from the model is saved in 'output' folder.

Libraries Used

• Keras : 2.2.0
• Tensorflow : 1.10.1
• Numpy : 1.14.5
• OpenCV : 3.3.0.10
• Scipy : 0.19.1
• Matplotlib : 2.0.2

Authors

• Dr Muhammad Atif Tahir (Supervisor)
• Hamza Usmani
• Ibad Saleem
• Abdul Rehman

Installation

Install Anaconda

https://www.anaconda.com/download/ (Make sure anazonda path variable is set)

Verify conda

conda --version

Clone the project

  git clone https://github.com/hamzausmani302/Domain-Specific-Anomaly-Detection-in-video-surveillance-using-transfer-learning.git

Go to the project directory

  cd Domain-Specific-Anomaly-Detection-in-video-surveillance-using-transfer-learning

Load conda environment and install dependencies

  conda env create -f testEnv.yml

Install nodejs from https://nodejs.org/download/release/v16.20.0

We have testig on node 16 other version will also work work , but install 16 for safe side.

   

 

(Optional but important GPU is required) The command above wiil setup cudatoolkit and cudnn but if GPU is still not detected, install cuda software and cudnn library based on the graphic card version. Link for CUDA setup:

https://medium.com/geekculture/install-cuda-and-cudnn-on-windows-linux-52d1501a8805

Check GPU is setup and detected

Usage/Examples

import tensorflow as tf

if tf.test.is_built_with_cuda() and tf.test.is_gpu_available():
    print("TensorFlow is running with GPU support.");
else:
    print("TensorFlow is running on CPU.");

or run the testModel.py file

Run Application

1. Move to root folder of the project in Anaconda Prompt (open Anaconda Prompt and navigate to the root folder)

2. Navigate to the GUI folder

    cd GUI
    npm install
    node index.js

Features

• running predictions on videos and extracting anomalous frames.
• running predictions on videos and generating graphical frame level predictions.
• running application on live feed
• Context Switching based on camera as well as day and night

Demo

Project Files

GUI/

Controllers/
public/
    assets/ - contain static data for websites
    Files/  - contains files uploaded by application
    Inter/  - Contains file after being broken down into sub files
    Outputs/  - outputs of the prediction on these files
    Views/ - contains html file and javascript code for the file
Routers/
    router.js  - api routes for the application
scripts/
serverUtils/
    cameraUtil.js  - camera functions
    fileUtils.js  - library for file related operations
    utils.js - general operations library
backupindex.js
config.js   -  application configuration
index.js  -   contains business logic for running predictions.. brain of application
package.json

NOTE_FRAMES/

framestream.py - display videos with frame number

preds_path/ - stores predictions from predict_test_set.py file for evaluating model

testFeatures/ - store test features after extration.

trained_models/ - store models

sports1M_weights_tf.h5
frontGate-day.mat

utils/ - general functions used by the python files for prediction.

c3d.py - C3D model file

classifer.py - contains the predictor network.

demoCustom2.py - script for graph video prediction used by index.js file

demoCustom3.py

demoCustom4.py - script for ouutputting frame image predictions used by index.js file

clip_maker.py

clip_maker2.py

clip_maker3.py - break videos into 30 seconds clips and save them in Inter/ Folder.

feature_extract.py - extract C3D features for prediction.

read_footage_cam.py - script used during live prediction.

temporal-annotations.txt - temporal file for evalurating metrics

predict_test_set.py - test the features and output predictions in pred_path folder

configuration.py - configuration for the python files

cameraConfig.json - centralized view of configuration of cameras

cameraConfig.py - retreiver functions from cameraConfig.json file for use by the application