Progress of Object Detection Based on Deep Learning

Object detection is commonly utilized in fundamental computer vision research fields: medical, autonomous driving, sensing monitoring, and other fields. With the advancement of high-performance computing technology, the computing speed of hardware has made a significant advancement. Meanwhile, machine learning-related algorithms, especially deep learning-based algorithms, greatly boosted the detection speed and accuracy. In practical applications, engineers have little understanding of the principles, effects, and performance of these algorithms, which seriously hinders the industrial application of object detection. However, the advancement has fueled challenges and competition for better modules and industrialization. Accordingly, the main purpose of this study is to present the algorithms of object detection and summarize it from three aspects: data set, algorithm, and performance, so as to provide a reference for researchers in related fields. In terms of algorithms, the paper mainly introduces anchor-free, single-stage, and two-stage algorithms. Finally, the existing problems and future research directions for target detection are discussed.