The pandemic changed the face visual surveillance thanks to a massive, mask-wearing public. It may be less common in the U.S. right now, but not around the world. While mask recognition software is available, New Mexico State University graduate student Anik Alvi pointed out that existing tools are proprietary, expensive, not compatible with other systems and therefore not widely used around the world.
Alvi decided to create his own algorithm to do the same thing at less cost with greater compatibility and available to all. He proposed the use of an autonomous surveillance system in his paper, titled “Semi-supervised, Neural Network based approaches to face mask and anomaly detection in surveillance networks,” published in the Journal of Network and Computer Applications.
“In this modern world, visual surveillance is present in almost every place,” Alvi said. “From malls to banks, offices to homes, every place is monitored with many cameras for security reasons. It becomes tedious and monotonous for a security officer to stare at multiple screens for a long period of time. This leads to many suspect activities going unnoticed. To avoid such incidents, there is a need for automated visual surveillance algorithms, which can detect any intrusion activity removing the burden of human security officers to constantly monitor.”
Alvi explained the system uses Multi-task Cascaded Convolutional Neural Networks (MTCNN) as the face detector, followed by a Gabor image feature extractor from a range of image extractors that identify key characteristics in an image and a Kernel-based Online Anomaly Detection (KOAD) algorithm for detecting and identifying the potential risk in real-time.
Developed in 2016, MTCNN is an algorithm in facial recognition technology that detects, aligns and extracts facial features from digital images.
The framework was tested on five different datasets comprising two datasets from public online repositories, which are online databases that allow research data to be preserved over time.
“We collected video data from surveillance cameras which we then converted to images at single timestep intervals. Within these images, we ran our proposed algorithm along with comparable algorithms to detect if a person is wearing face mask or not. Our proposed algorithm performed better than the comparable algorithms, at an accuracy of 78%.”
Born in Dhaka, Bangladesh, the computer science graduate student, started working on this project as an undergraduate research assistant and led to Alvi publishing his first conference paper “Comparative Study of Traditional Techniques for Unsupervised Autonomous Intruder Detection.”
Alvi came to NMSU to finish his research after receiving the NMSU International Competitive Scholarship to pursue a master’s in computer science. He currently works as a graduate assistant at the New Mexico Water Resources Research Institute (WRRI) and has continued other research while earning his master’s degree.
“I have been doing research related to Machine Learning and Human-Computer Interaction with researchers from different universities around the world. I have also published papers in leading conferences and journals which includes the Institute of Electrical and Electronics Engineers, Springer and Elsevier.”
Alvi is now currently looking for new collaborative projects related to machine learning to improve public and environmental safety, a goal that greatly influenced his most recent work.
“This system can assist in public safety by enforcing safety protocols in various settings, such as airports, hospitals, public transport and crowded spaces, contributing to ongoing efforts to prevent the spread of infectious diseases,” Alvi said. “Also, the ability of this system to identify irregular activities or behaviors can aid in preventing crimes, tracking suspicious movement and enhancing overall security measures in public and private spaces.”
The full article can be seen at https://newsroom.nmsu.edu/news/nmsu-researcher-s-paper-proposes-open-source-autonomous-surveillance-systems/s/a5e49182-d086-483d-baa6-dff654a6cced