A survey of deep learning techniques for detecting and recognizing objects in complex environments

• Published in: Computer science review Vol. 54; p. 100686
• Main Authors: Dogra, Ashish Kumar, Sharma, Vipal, Sohal, Harsh
• Format: Journal Article
• Language: English
• Published: Elsevier Inc 01.11.2024
• Subjects: Classification; Complex environment; Convolutional neural networks; Localization; One-stage detector; Two-stage detector; One-stage detector; Two-stage detector; Convolutional neural networks; Localization; Complex environment; Classification
• ISSN: 1574-0137
• DOI: 10.1016/j.cosrev.2024.100686

Object detection has been used extensively in daily life, and in computer vision, this sub-field is highly significant and challenging. The field of object detection has been transformed by deep learning. Deep learning-based methods have shown to be remarkably effective at identifying and localizing objects in images and video streams when it comes to object detection. Deep learning algorithms can precisely locate and localize objects inside photos and videos because of their capacity to learn complex and nonlinear patterns in data. Deep learning models may also be trained on big datasets with minimal human intervention, allowing them to rapidly improve their performance. This makes deep learning models useful for applications such as self-driving cars, recognizing faces, and healthcare diagnosis. The purpose of this study was to gain an in-depth understanding of the primary state of development for the object detection pipeline in complex environments. Initially, this study describes the benchmark datasets and analyzes the typical detection model, and then, the paper systematic approach covers both one-stage and two-stage detectors, giving a thorough overview of object detection techniques in complex environments. We also discuss the new and traditional applications of object detection. In the end, the study reviews how well various topologies perform over a range of parameters. The study has covered a total of 119 articles, of which 27% are related to one-stage detectors, 26% to two-stage detectors, 24% to supporting data related to deep learning, 14% to survey articles, 8% to the datasets covered in the study, and the remaining 1% to the book chapters.