EXAMINATION OF THE CORE TECHNOLOGIES FOR PEDESTRIAN DETECTION THROUGH DEEP LEARNING.

Abstract

Visual perception is necessary for the efficient operation of various different kinds of intelligent technology. There are many instances of objects that fall within this category, such as autonomous vehicles, robots, and surveillance systems. Even while deep learning is still a relatively new discipline, it has already made a significant amount of headway. In comparison to earlier computer vision models, deep architectures, which include convolutional neural networks (CNNs), are superior. Deep learning significantly simplifies the process of identifying pedestrians. This article aims to comprehensively analyse the available research and approaches for pedestrian identification. This article examines a few cutting-edge models that are considered the greatest in their respective disciplines. A CNN that is composed of regions is an example of a model that satisfies this particular set of requirements. CNNs such as the R-CNN, Fast R-CNN, and Faster R-CNN are all examples of the types of networks that fall under this category. Another sort of detector is the transformer-based variant, and further types include the single-shot detector (SSD, YOLO) and the transformer-based variation. The study encompasses techniques for data preparation, data addition, and data processing modifications, including non-maximum suppression. This research also includes methods for modifying processed data. Researchers will discuss each of these facets in detail throughout the presentation. When researchers examine the performance of different systems based on a variety of parameters, such as recall, average miss rate, and accuracy, researchers may be able to obtain a great deal of insight into how these systems operate.