Firearms Detecting

Project Title

Firearms Detecting

Project Area of Specialization Artificial IntelligenceProject Summary

The rapid advancement in the domain of machine learning has revolutionized the way humans perceive multi-dimensional data. There is an immense need to develop solutions for the betterment of mankind using data (especially big data). For this purpose, we want to detect hidden guns or firearms with the help of thermal cameras and deep learning solutions. This will increase the sense of security and help law enforcement agencies to reduce criminal activities. Many criminal acts like robbery, theft, and terrorism are commonplace these days. Law enforcement agencies have taken different measures to reduce the unethical use of firearms by installing metal detectors at various places. A similar approach to alleviating the use of firearms for false purposes is the automatic detection of hidden firearms.

We are focused on detecting hidden firearms using state of the art deep learning methods. Some recent work has been done in detecting and tagging visible firearms, however wide variations in the shapes, sizes and the angles it is captured from makes it hard to detect. Besides this, obstruction is another factor which slows down the progress of detecting hidden firearms. We will be using heat vision cameras to detect the hidden firearms, as the camera captures the images/videos on the basis of the heat content; the resultant images would have the heat content from the person’s body with the blockage of the object he/she is hiding inside, thus forming an outline of the object. The next step is to apply deep learning based computer vision techniques to identify the object. Ultimately, we train our model on the acquired dataset, implement algorithm including YOLO and its variants to perform real-time object detection.  

Project Objectives

One of the objectives includes a thorough literature review on firearms and object detection that help us out for implementing the detection algorithm based on scientific research. The primary objective is to assist law enforcement agencies in detecting hidden firearms, to develop a real-time solution. This will increase the sense of security in common people and ensure security and surveillance with the help of state of the art technologies.

Project Implementation Method

For the collection of our dataset, we require high-end thermal cameras which operate in the non-visible light spectrum to perceive hidden objects. For this, we will contact different shooting clubs and law enforcement agencies to capture images of concealed guns by using a thermal camera. On visible firearm detection, work has been done recently and we will extend our work to hidden firearm detection and make it real time. The proposed methodology includes implementing the state of the art object detector You Only Look Once (YOLO) algorithm incorporating its recent version YOLOv3 for real-time multi-object detection. YOLO is based on deep learning, has real-time speed to detect objects. This includes recognizing an object using deep learning framework, that will compute the strong features using convolutional layers. These features will then classified by using a fully connected network.

Reference papers:

Redmon, Joseph, and Ali Farhadi. "Yolov3: An incremental improvement." arXiv preprint arXiv:1804.02767 (2018).

Iqbal, Javed, Muhammad Akhtar Munir, Arif Mahmood, Afsheen Rafaqat Ali, and Mohsen Ali. "Orientation Aware Object Detection with Application to Firearms." arXiv preprint arXiv:1904.10032 (2019).

 

Final Deliverable

Our final deliverable includes a complete software demo and prototype for hidden firearms detection. With the help of funding, we will be able to purchase a thermal camera and to make the dataset. This dataset will be processed using a deep learning algorithm to detect objects. Future directions of our work will include a hardware-based implementation of object (firearms) detection.

Benefits of the Project

The successful implementation of our project helps to reduce robberies, theft, and terrorism. Law enforcement agencies will benefit from the proposed solution, help to increase the sense of security as the crime rate will be decreased. In our country, this issue needs to be addressed as the law-and-order situation demands an efficient solution for controlling events like target killings, robberies, and terrorism. Our solution will help to mitigate such activities to a noticeable extent. Assailant usually carries the gun (pistol, revolver, semi-automatic pistol) inside the jackets and blazers, and pointing out such people becomes difficult in public places. Hence, developing the solution to detect concealed guns can help law enforcement agencies to point out a mugger by notifying the concerned authorities. Moreover, it will also help in reducing medical budgets by stopping potential violence before its occurrence and can save precious lives and money.

Technical Details of Final Deliverable

We are focused on detecting hidden firearms using state of the art deep learning methods. Some recent work has been done in detecting and tagging visible firearms, however wide variations in the shapes, sizes and the angles it is captured from makes it hard to detect. Besides this, obstruction is another factor which slows down the progress of detecting hidden firearms. We will be using heat vision cameras to detect the hidden firearms, as the camera captures the images/videos on the basis of the heat content; the resultant images would have the heat content from the person’s body with the blockage of the object he/she is hiding inside, thus forming an outline of the object. The next step is to apply deep learning based computer vision techniques to identify the object. Ultimately, we train our model on the acquired dataset, implement algorithm including YOLO and its variants to perform real-time object detection.  For the collection of our dataset, we require high-end thermal cameras which operate in the non-visible light spectrum to perceive hidden objects. For this, we will contact different shooting clubs and law enforcement agencies to capture images of concealed guns by using a thermal camera. On visible firearm detection, work has been done recently and we will extend our work to hidden firearm detection and make it real time. The proposed methodology includes implementing the state of the art object detector You Only Look Once (YOLO) algorithm incorporating its recent version YOLOv3 for real-time multi-object detection. YOLO is based on deep learning, has real-time speed to detect objects. This includes recognizing an object using deep learning framework, that will compute the strong features using convolutional layers. These features will then classified by using a fully connected network.

Final Deliverable of the Project Software SystemType of Industry IT , Security Technologies Artificial Intelligence(AI), OthersSustainable Development Goals Sustainable Cities and CommunitiesRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	65000
Thermal Camera	Equipment	1	65000	65000