A survey on the modern research on deep learning-based object detection

Detecting objects remains a vital and challenging aspect within computer vision applications. Substantial progress in detecting objects is made by enhancing how objects are represented and by utilizing deep models of neural networks. This paper delivers a broad survey on the advancements on object detection within the deep learning era in recent years. Object detection algorithms are generally branded into two broad types centered on their training approach: single-stage detection algorithms and two-stage detection algorithms, each utilizing unique methods for pinpointing objects in images. We examine the underlying rationale of these algorithms by conducting experiments to compare various aspects, including quality metrics, trade-offs between speed/accuracy, and modelling practices. This study also provides a comparative assessment of leading convolutional neural networks (CNNs) used in detecting objects, emphasizing each model’s strengths/limits and presenting key conclusions. To foster a deeper understanding, we also identify areas for future research and potential advancements in object detection models within the domain of deep learning.