This chapter explains various facial identification algorithms used in drones. The burgeoning field of face recognition technology makes use of image processing to identify faces in people. Face recognition is becoming increasingly popular for a variety of reasons, such as the growing population, which necessitates higher security and surveillance systems, identity verification in the digital age, combat in rural regions, disaster relief, and so forth. This research compares and contrasts various face identification techniques, including neural networks, PCA, LBPH (local binary pattern histogram), PAL, capsule network, and LDA (linear discriminant analysis). Additionally, a comparison of face datasets collected with drones is included in this chapter. The results of this study will help facial recognition technologists design a hybrid algorithm that meets the requirements of real-time applications.
TopIntroduction
Drones, or unmanned aerial vehicles (UAVs), are a crucial component of military and public security scenarios in a world where video surveillance has become a crucial aspect. Nowadays, everyone can afford and operate these vehicles due to their ease of use (Hosni et al., 2019). The fact that they are nearly imperceptible to radar and are able to transport a payload such as an optical zoom high-resolution camera is a major advantage. Facial recognition is one of the most popular aerial surveillance applications, partly due to the significant advancements in deep learning that have fueled this field of study (Ullah et al., 2019).
The face is the essential component of an individual that is mostly utilized to identify them. Even if people are able to identify familiar faces, it gets harder and harder to identify unfamiliar faces. This is the point at which an automated system that can recognize people just as well as humans did was created. Nowadays, face recognition technology is widely employed in practically every aspect of life (Anwar et al., 2020; Rahim et al., 2023). Face recognition is being used in real-time in a variety of industries, including law enforcement, immigration checks, forensic investigations, attendance tracking, disease diagnosis, and more (Hosni et al., 2018). The demand for items with additional features has increased due to these rising technologies. As a result, technological fusion developed (Haq et al., 2022). But there are different benefits and drawbacks associated with each algorithm. To make the most of face recognition technology, this research has integrated it with drone technology (Haq et al., 2024). Different machine learning and deep learning algorithms can be used to create face recognition drones, but each approach has pros and cons of its own.
There are three main steps in the facial recognition system's identification process (Ahmad et al., 2022). As seen in Fig. 1, they are Acquisition, Extraction, and Recognition. Acquisition is the process of gathering and organizing photographs of people taken from various viewpoints, lights, and expressions into a database. The process of extraction involves gathering distinct facial traits, such as cheekbones, jaw line length, and eye spacing, in order to compare and identify each one (Siddiqui et al., 2023). By comparing the individual face with the faces in the database, the match is discovered (Fatima et al., 2022).
This research aims to examine various algorithms and datasets that have nearly resolved the predetermined parameters influencing recognition accuracy and weigh the advantages and disadvantages of each approach.
TopAlgorithms For Drone Face Recognition
Face identification for photos taken by drones poses special difficulties because of differences in resolutions, positions, and illumination. Drone face photos have led to the adaptation or creation of numerous face recognition algorithms.
Eigen Faces
It involves calculating the principal components and applying them to alter the data; occasionally, only the first few principal components are used, with the remaining principal components being ignored (Kshirsagar et al., 2011). It is a dimensionality reduction approach that minimizes the amount of variance in the dataset while reducing it from a larger set to a smaller set of variables. In machine learning, this algorithm is an unsupervised learning algorithm. This algorithm's main constituents are Eigen values, Eigen vectors, variance, and covariance. Obtaining the dataset and splitting it into a training set and a validation set is the first stage. Then, a two-dimensional matrix with rows representing data items and columns representing features should be used to represent these datasets. Eigen values and Eigen vectors are calculated using the covariance of the matrix, which is discovered after the data has been standardized. These eigenvectors are ordered from large to small, and the irrelevant features are eliminated from the dataset.