A Survey on Computer Vision Methods and Approaches for the Detection of Humans in Video Surveillance Systems

Today’s societies require special attention to control human access in security-sensitive areas. Traditional surveillance methods demand significant manpower and financial resources, and monitoring multiple video screens simultaneously can be a burdensome task for human operators. Therefore, the development of an automated video surveillance system that can detect unauthorized human intrusions and alert security personnel are crucial. Automated surveillance systems based models for feature extraction delve into the classification of human and non-human images using various machine learning and deep learning techniques such as RPN, CNN, ANN, DNN, Adaboost, and SVM. Additionally, the paper conducts a detailed examination of the implementation of computer algorithms for real-time human detection on hardware platforms, including Stayton, Xilinx, Virtex, GPU, and FPGA processors. Numerous factors can impact the accuracy of such systems, including variations in environmental conditions, human body shapes, poses, and backgrounds. The paper briefly explores the challenges associated with each of these processes. It concludes by summarizing the methods discussed and highlighting the most effective algorithms. This summary aims to provide researchers in this field with insights into the scope and challenges associated with the automatic detection and classification of humans. This paper aims to provide a comparative study of recent advancements in image processing for automating human detection across various applications. The paper discusses algorithms and techniques employed in image processing, covering database collection, preprocessing methods for foreground extraction, and extraction of features. It explores both handcrafted models and deep learning.