Research Advanced in 2D Multi-Object Tracking based on Deep Learning

Abstract

Multi-object tracking is one of the most important research orientation of computer vision, which has a broad applications in the fields of sports game data analysis, video monitoring, pedestrian behavior analysis, and automatic driving. With the rapid-developed convolutional neural network, deep learning-based multi-object tracking methods have made significant progress. In this research, the current multi-object tracking methods are classified, analyzed and summarized through comprehensive and thorough research. Firstly, the definition of multi-object tracking, research background and application advantages of deep learning are introduced. Next, 2D multi-object tracking and 3D multi-object tracking as well as their differences are briefly introduced. Then, different methods in 2D multi-object tracking are explored in detail. The commonly used evaluation metrics and datasets are also introduced, as well as the comparison of the experimental results of different methods on the datasets. Finally, the advantages and disadvantages of multi-object tracking methods are analyzed and the future development direction is prospected. The development of multi-object tracking is of great significance for enhancing social security as well as promoting technological progress.