Rare Event Detection Using Deep Learning Algorithm

Surveillance cameras have become a part of our life. We see them on the corner of streets outside restaurants and shops and different places. However, hiring people to monitor all the surveillance video is expensive. The purpose of our project is to implement an algorithm that will detect ab

Project Title

Rare Event Detection Using Deep Learning Algorithm

Project Area of Specialization

Artificial Intelligence

Project Summary

Surveillance cameras have become a part of our life. We see them on the corner of streets outside
restaurants and shops and different places. However, hiring people to monitor all the surveillance
video is expensive. The purpose of our project is to implement an algorithm that will detect abnormal
activity such as robbery, accident, etc. and alert the corresponding authorities. We will design a
classification method to classify the event. After that, we will design an alert system to alert the
corresponding person or authority on the rare or abnormal activity took place.
We will be using Deep Learning techniques with CNN (Convolutional Neural Network) model to detect
rare events for example, a vehicle on a footpath, robbery etc. Deep learning is a subset within Artificial
Intelligence where deep neural networks are trained on a large amount of data. We will divide our
videos which are anomaly positive in positive box and anomaly negative in negative box then we will
divide each video into temporal segments and then extract Convolutionary 3D feature
which is great for event detection in videos.

Project Objectives

As the government does not have the capacity to monitor surveillance feeds. Moreover, surveillance
video monitoring is waste of time and waste of labor when it can be done by the proposed model. Even,
corrupted person hired to monitor the surveillance the video can also abuse his/her access.

Project Implementation Method

Uptil now we have done the first phase of our project by using tensorflow library and implementing the approach of Multiple Instance Learning on the online platform "Google Colab". Multiple Instance Learning is an approach where we are actually dividing the video into frames and creating small parts (16 images per segment) and putting 32 segments in one temporal bag which creates it easy to detect the anomaly happening in the video at any point of the time. After completing the 1st phase of the project now we are facing difficulty to work on big dataset and updating our project from Command Line Interface towards Graphical User Interface for that purpose we need a GPU Cluster machine and any kind of access to it whether it can be a VPS or a physical machine.

Benefits of the Project

As the government does not have the capacity to monitor surveillance feeds. Moreover, surveillance
video monitoring is waste of time and waste of labor when it can be done by the proposed model. Even,
corrupted person hired to monitor the surveillance the video can also abuse his/her access.

Technical Details of Final Deliverable

• Dataset Collection
• Pre-processing of videos
• Successful Implementation of the model by using Multiple Instance Learning approach
• Make predictions using the model and compare the results
• Updating the Command Line Interface to Graphical User Interface
• Working on Pakistani Dataset

Final Deliverable of the Project

Software System

Core Industry

IT

Other Industries

Security

Core Technology

Artificial Intelligence(AI)

Other Technologies

Others

Sustainable Development Goals

Decent Work and Economic Growth, Industry, Innovation and Infrastructure, Sustainable Cities and Communities, Peace and Justice Strong Institutions

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com