Malicious UAV Detection Using Integrated Audio and Visual Features for Public Safety Applications.

Area of Science:

• Computer Science
• Electrical Engineering
• Security Studies

Background:

• Unmanned aerial vehicles (UAVs) are increasingly used for surveillance but pose significant privacy risks.
• Timely detection of malicious drones remains a critical challenge for security firms.
• Existing drone detection methods have limitations, including dataset size and environmental factors.

Purpose of the Study:

• To propose a novel framework for detecting and localizing malicious drones.
• To address the limitations of current drone detection schemes.
• To enhance security against unauthorized aerial surveillance.

Main Methods:

• Developed a hybrid framework integrating handcrafted and deep features.
• Utilized both sound and image data for drone detection.
• Incorporated datasets with occluded images and varied environmental conditions (resolution, illumination).
• Employed Support Vector Machine (SVM) with various kernels for feature classification.

Main Results:

• The proposed hybrid framework demonstrated improved performance in detecting and localizing malicious drones.
• The method effectively utilized combined acoustic and visual information.
• Experimental results showed superior performance compared to existing related methods.

Conclusions:

• The novel hybrid feature framework offers a promising solution for malicious drone detection.
• The approach is robust to variations in image quality and environmental conditions.
• This research contributes to advancing security measures against drone threats.