Identifying Suspicious Intruder utilizing Depth Images and Machine Learning

The safety and security of human beings have always been of the utmost concern in every aspect. With the advancement in engineering, sciences, and technology new methods are been introduced to harm human beings. Over the past several decades, one such nuisance is the Suicide Bombing that is still an

Project Title

Identifying Suspicious Intruder utilizing Depth Images and Machine Learning

Project Area of Specialization

Artificial Intelligence

Project Summary

The safety and security of human beings have always been of the utmost concern in every aspect. With the advancement in engineering, sciences, and technology new methods are been introduced to harm human beings. Over the past several decades, one such nuisance is the Suicide Bombing that is still an open challenge for the world to predict. Our project deals with the identification of a suicide bomber using a 3D Depth Camera and Machine Learning techniques. It utilizes the skeletal data provided by a 3D Depth Camera to identify and alarm us whether a subject is a suicide bomber or not. The prediction is based on real-time 3D posture data of the body joints obtained using a Depth Camera sensor. This project extracts 3D spatial features of the human body for 20 joints. Initially, using a comprehensive experimental design we have created a dataset from scratch consisting of 3D spatial gait information of 20 joints of various subjects with and without a suicide jacket using Kinect v1. Additionally, the dataset will also contains the subject’s height and weight. Utilizing this dataset, we then applied different machine learning classification algorithms to train a model that gave us a higher accuracy in predicting the attached labels in the dataset to normal subjects and suicide bombers. Experiments were performed with the suicide jacket bearing 8 to 20 kg weight. Finally, the model with the highest accuracy was implemented inside our Live Detection prototype which detects a suicide bomber in real-time using the 3D Depth camera and alerts the necessary authorities from a graphical point of view.

Project Objectives

Our project summarized here aims to address the issue of identifying a suicide bomber from a good distance using a 3D Depth camera and machine learning methods. The key purpose of this project is to address the human society issue by focusing on the identification of suicide bombers before they could execute an attack and harm other human beings. The main objective of our project is to study for skeletal deflection in joints of a normal subject and a one wearing a suicide vest. Height and weight data captured will additionally allow us to categorize a person into six different categories. The same data will then be used as a ground truth for the machine learning model to infer. After studying, we aim to develop a live detection system that can in the future be used as a security and surveillance system. The objective of this system would be to detect a suicide bomber from a crowd enabling the respective authorities to be alerted.

Project Implementation Method

As the suicide attacks are on the rise, human security has become a major concern. It was essential to develop a method that can detect a suicide bomber, with some distance and quickly alarm the concerned departments. This was achieved by identifying a suicide bomber using a 3D Depth Camera and Machine Learning Algorithms. The experimental setup utilizes the 3D Depth Camera as the RGBD camera for data acquisition. To mimic the actual behavior of a suicide bomber, a synthetic suicide vest taken with a capacity to carry explosives of varying weights (08-20 kg). The weights were adjusted inside the jacket according to the conventional placement in real-world scenarios, i.e. ribs area, chest, and backbone area. During experimentation, the subjects were asked to walk in front of the 3D depth camera sensor with and without wearing the jacket in this case the weights were varied from 08-20 kg with an interval of 04 kg. Data collected for every subject consisted of X, Y, and Z coordinates for each of the 20 joints. The height of a person was calculated using these coordinates and the weight of each subject with and without a vest was also determined using a digital scale balance. Each individual has a different walking posture and it is rare to find multiple individuals with exactly the same walking style. However, even having an identical walking posture of two or more persons makes it difficult to map them because of the difference in weight and height. These two factors directly affect the walking posture of any person. For this, each subject was categorized into one of the six sizes; X-Small, Small, Medium, Large, X-Large, XX-Large. This categorization is based on the subject’s weight captured and height estimated. In addition to this, we utilized four classifiers to predict the vulnerability of an individual to be a suicide bomber. These were Support Vector Machine (SVM), k-Nearest Neighbor (k-NN), Random Forrest (RF), and Artificial Neural Network (ANN). The experiments were performed to compare the performance of the classifiers in terms of accuracy, sensitivity, specificity, recall, and precision. The Machine Learning Classifier on which the maximum accuracy was obtained on both training and testing dataset was taken into consideration. Furthermore, the GUI application was designed from scratch in Python language utilizing PyKinect (Kinect v1) and PyGame libraries.

Benefits of the Project

In the past, many systems like Backscatter x-ray scanner, Millimeter Wave Scanner, Walk-through metal detector have been implemented outside the areas which need high security but either these systems had some negative impact to the human health or they were banned in many countries or had less accuracy in identifying suspicious intruder. Amongst the many models which identify suspicious intruders in the modern world, doing this task through depth images with machine learning classifiers remain the modern focus for many engineers and scientists till date since this proposed solution is much better in accuracy and can detect a suicide bomber from the distance of about 0.2 – 20 meters using Stereolabs Zed 2 Stereo camera. In this research, the data showed the deflection in joints of a normal human skeleton with and without a suicide vest. This system can be implemented in the areas where high security is needed like shopping malls, Hospitals, Banks, and etc. where the depth camera can be installed at the top of the entrance of the gates which will automatically ring an alarm when the depth camera identifies a suspicious intruder entering that particular vicinity. Moreover, it can also be installed at the entrance of Offices, Government and Army Offices, Airports, Educational Sectors, and etc. The key challenge in today’s suicide attacks is the identification of a suspected bomber from a good amount of distance which can be surely done using Stereolabs Zed 2 Stereo camera in place of a Kinect v1 which has a depth Range of 0.2 – 20 m. Many countries have significantly suffered due to this problem, some of them include Afghanistan, Pakistan, Syria, and Yemen. Hence, the proposed solution can prove out to be much higher in accuracy and a modern approach to work with.

Technical Details of Final Deliverable

Since we aim to detect a suicide bomber from a good amount of range, hence our final deliverable is a Graphical User Interface (GUI) based application that can readily be installed along with the necessary hardware to monitor for a suicide bomber in a particular area or enclosure. The GUI Application is made from scratch in Python using the PyKinect and PyGame libraries. PyKinect library enabled us to work for now with the Kinect v1 which has a low-cost 3D Depth camera but has a restricted range of 4 - 4.5 meters. Using these libraries we developed a GUI that extracts data in real-time for up to 6 people at a time. New depth cameras for e.g. Stereolabs Zed 2 Stereo camera can detect even more subjects at a particular time. Using the data collected, we pre-processed it by adding along with the height, weight, and category of the subject. We then loaded the pre-loaded AI model at the backend and forwarded the data to the model which inferred the suicide bomber. If a particular subject is one, his skeleton is updated in red color and a buzzer is sounded at the back which alarms the respective people or authorities. A better depth camera would enable us to detect a subject from a good amount of distance and would also help initially since good depth cameras can detect more skeletal joints, as compared to a Kinect v1. Additionally, they can also track more subjects in real-time too.

Final Deliverable of the Project

HW/SW integrated system

Core Industry

Security

Other Industries

Education , IT

Core Technology

Artificial Intelligence(AI)

Other Technologies

Sustainable Development Goals

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

Required Resources

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