Assault detection through video surveillance systems

The increase in the rate of crimes like harassment and assault, nowadays, has become a very serious issue. According to the reports, most of the women go through verbal harassment, about 71%. In workplaces, 27% of women and are being sexually harassed and assaulted. 41% of female health care workers

Project Title

Assault detection through video surveillance systems

Project Area of Specialization

Artificial Intelligence

Project Summary

The increase in the rate of crimes like harassment and assault, nowadays, has become a very serious issue. According to the reports, most of the women go through verbal harassment, about 71%. In workplaces, 27% of women and are being sexually harassed and assaulted. 41% of female health care workers reported experiencing some form of violence. In public places, 51% of women are used to facing unwelcome and unethical touch. On the other hand, children are also being assaulted, physically as well as sexually. In the past few years, many cases of children being assaulted in educational and religious institutions have emerged.

 To alleviate the rapid increase of these types of cases, we propose a real-time surveillance system that will detect any kind of physical and verbal assault by using advanced deep learning algorithms.

CCTV cameras have been playing an important part in the security of the cities, but due to the increase in the number of CCTV cameras than the operators, it has become quite difficult to manage the system. Our project will take over this problem, by alerting if any kind of violence has occurred. The camera will send the data to the computer where the first object detector will detect the human presence than "situation of interest" is continuously being checked by deep learning algorithms and it will recognize the voice of the activity is verbally being done. The CCTV camera will then trigger an alert which will be sent to the concerned authorities.

Another invigorating point of our project is privacy and confidentiality. In places where privacy is important like offices and institutes, CCTV's recording will only be active at the time when the system will recognize some keywords which will be given in the dataset.

Additionally, we are developing a database for recording all the anomalies to carry effect actions against the assaulter.

Fig 1: The Conceptual Model

Project Objectives

We aim to propose a high computational information processing surveillance system employed in 1) work environment and 2) in educational institutes, for automated identification of the assault or violence.

We shall define a “situation of interest” as an unwelcomed and undesirable situation, like assault and violence along with verbal assault. We are using the assault dataset at the beginning of our project, later we aim to include harassment, bullying, and abuse. The techniques which will detect the “situation of interest” are based on the more accurate and speedy deep learning algorithms.

The intelligent surveillance system comprises of few night-vision 1080p CCTV with two-way audio which will be easily installed at the workplace and educational institutes without buying any hardware.

• To find datasets on the internet, and custom train our assault model.
• For real-time detection, we are using a deep learning algorithm having higher accuracy as well as speed.

• To optimize the memory and processor, the camera will not record in the normal situation. It will only get active when there are at least two humans detected and will only record when finding physical or verbal anomaly.
• To make such a user-friendly smart surveillance system that can be used by anyone which eliminates the need for a technical instructor.
• To create a database where the date, time, face, gender, and other details of the assaulter and victims will be stored. The database can only be accessed by concerned authorities. The data can be further used for other projects if granted.

Project Implementation Method

We aim to train the dataset available on the internet on assault through the accurate and speedy model to apply on the real-time environment for real-time Assault Detection in a) work environment and b) educational institutes.

The flow chart given below explains well the methodology we are going to follow.

As CCTV cameras are mostly implanted in most places, the CCTV footage is being monitored on some screens. Humans are prone to error. Neither every abnormal situation gets noticed nor our devices can record every frame to identify the crime person later. Therefore, we are going to apply some advanced deep learning algorithms which are more accurate than a human, who never gets tired of executing each frame.

Each frame of CCTV footage will be executed. First, it will search for human through an Object detector algorithm, neither any other algorithm will be executed, nor footage is recorded until more than one human is detected. The assault detection algorithm is not applied in normal activities. When any anomaly occurs only then our algorithm plays its role to detect assault. It then starts recording those frames along with identifying the face of the assaulter to store it in the database (time, location, gender, and face). The concerned authorities are also alerted through notification.

As there can be a verbal bully too, the audio of CCTV footage is also being checked for bullying words.

Fig 2: Flowchart

Benefits of the Project

The increase in the number of cases of assault has become quite alarming. Nowadays, the cities are being monitored by CCTV cameras but detecting each interpersonal crime is only limited, as there are several CCTV cameras than some operators. So, it is nearly impossible to detect the crime and trace the location in real-time. If someone is familiar with image classifier, he knows that the deep learning algorithms have taken over human accuracy.

Our project will detect the crime and it will be triggered when a “situation of interest” will occur. The response of the trigger could be an alert to be issued to the concerned authorities. In workplaces, educational and religious institutes, where privacy is important, but the cases are used to occurring, this validates the requirement of the smart surveillance system which can ensure privacy as well as confidentiality.

Our project will be providing privacy by responding only to such triggers if the situation is of our interest, otherwise, it will not record the voice at all. Hence, confidentiality and privacy are assured.

Many individuals of any age or gender face bully and hurtful words and behavior which might give a lifetime trauma. Victims of this undesirable behavior must be going to physical and mental health specialists at a later age. We can save them from lifetime depression by acting against the assaulters. This is will be a contribution to the health sectors in a way that people would not ignore this serious issue due to lack of time or money.

To save the processing time and device storage we are trying to find out the best algorithm as there are many comparisons of algorithms based on accuracy and speed. Our project will be time-efficient, technically feasible as well as easily available.

Our project has room for innovation. We are only focusing on detecting assault but as we get succeeded in detecting assault, we can also detect violence, harassment, bullying, and abuse, which is the future work for this project.

Technical Details of Final Deliverable

The final product is an intelligent, smart, and real-time surveillance system. We are not building new circuitry as this will be expensive to install everywhere.

As CCTV cameras are mostly cheap to install and they are being used for security purposes. We are applying some advanced deep learning algorithms on existing hardware along with smart storage and accuracy techniques. For verbal assault, we are using a CCTV camera that already has audio detection therefore by using a voice recognition algorithm we will be able to detect verbal assault too.

We aim to design a user-friendly interface so that technical operators are not needed to be employed.

The time, date, face, and other details of the crime person will also be recorded to share with concerned authorities in the database.

Final Deliverable of the Project

HW/SW integrated system

Core Industry

Security

Other Industries

Education , Medical , Health

Core Technology

Artificial Intelligence(AI)

Other Technologies

Cloud Infrastructure, Big Data

Sustainable Development Goals

Good Health and Well-Being for People, Quality Education, Industry, Innovation and Infrastructure

Required Resources

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