The Research And Implementation Of The Multi-target Tracking System Based On TLD Model

With the development of computer vision and artificial intelligence,target tracking technology in video gradually becomes a popular research topic,and it is widely used,such as the location and analysis of vehicles and pedestrians in intelligent transportation systems,action recognition in human computer interaction.This paper focus on the popular long time target tracking model,TLD,analyses the principle and procedure of the algorithm,and then extend it to multi-target tracking.Based on improved algorithm,we design and develop a stable target tracking system.The main work is as follows:1.Introducing and analyzing the long term tracking model(TLD),illustrating the principle and framework of the algorithm from three aspects: tracking,detecting and learning Module.We also compare TLD model with other tracking algorithm and give brief result of the advantages and disadvantages of TLD model.2.To overcome the two issues in TLD model: expensive time-consuming in scanning widows and single object tracking,we propose some solutions to extend the system to multi-object tracking.We design a mechanism to reduce the number of scanning windows.For multi-target tracking,we assign an identification ID to each target in the initialization step and use this unique ID to distinguish different targets in the following tracking,detecting and learning procedure.3.Designing and implementing a multi-target tracking system based on TLD model.This system is developed on MFC framework and Opencv Image processing library.Basic system interactions involve video reading,multi-target tracking and the saving of coordinate of target and video screenshot.We test the system on the video tracking benchmark database,the result shows that the system is effective in real-time multi-target tracking and its reliability and stability are satisfactory.