Chinese Journal of Computers   Full Text
Title	Human Tracking in Infrared Images Based on Particles Mean-Shift Migration Algorithm
Authors	YUN Ting-Jin GUO Yong-Cai GAO Chao
Address	(Key Laboratory of Optoelectronic Technology and Systems of the Education Ministry of China, Chongqing University, Chongqing 400030)
Year	2009
Issue	No.6(1222—1228)
Abstract &   Background	Abstract A novel method based on particles Mean Shift migration process for human tracking in infrared images is proposed in this paper. Reference to the basic particle filtering method, the core idea of this algorithm is to establish the object’s state model and measurement model dynamically by the migrating and clustering of random sampling particles. The sampling particles are placed on the tracking window randomly and a Mean Shift converge process is implemented by taking the particle’s corresponding pixels’ intensity as feature. The object’s state model is expressed by the convergent particle sets. Then, another Mean Shift clustering process is carried out on the particles sets of the state model using their coordinates in the image as features. The clustering results are regarded as the measurement of the object. In the sequential tracking process, the sampling particles in the next frame are generated randomly based on its previous frame measurement results. Comparing with the traditional Sequential Importance Sampling (SIS) based particle filter, the tracking method doesn’t need the object’s likelihood criterion, which is a big problem for infrared human tracking, and the robust tracking process can be achieved only by using a few sampling particles. Keywords particles migration; human tracking; infrared image; Mean Shift; particle filtering Background Infrared-image,especially the far-infrared image has prominent advantage comparing with the visible-image. As a result of thermal imaging, infrared image is independent of external luminous qualification and is able to see the interested objects in dark and frog environments, which is almost impossible in visible optical imaging. Human detection and tracking are very important issues which can provide the most active and valuable information in many occasions. The systems based on infrared imaging can almost work in any environment and all-weather conditions. It is irreplaceable in some situations and has widely potential applications in many aspects, such as the Infrared Navigation System, the Infrared Life-Saving System, the Frontier Defense Precaution System, the Smart Surveillance System, the Night Driver-Assistance System, the Fire Rescue and Public Safety System, the Man-Machine Interface System and the Robot Vision System, etc. However, human detection and tracking in infrared images is a challengeable task from the technical viewpoint, dealing with many fields of knowledge, such as image segmentation, feature extraction, pattern recognition, object tracking etc. Since human are no-rigid complex objects, the main difficulties we have to confront firstly is the un-reliability of human detection results in infrared image, caused by human’s various appearance and the intrinsic properties of infrared image. Because of infrared image is gray image, no color information is available and there are some other drawbacks such as image blur, low textures. With complicated and changeable postures & appearances, it is very difficult to extract and descript human features in infrared image effectively and to distinguish them from disturbances. Secondly, human’s movements are very subjective and unbending; there are no routines to recapitulate them. Meanwhile, the movements also accompany with human postures and appearances change, the tracking methods used in rigid objects are not suitable for human tracking. Thirdly, being short of features for tracking, some excellent human tracking algorithms based on color information and textures can’t work efficiently for infrared human tracking. Currently, the main research on this field is focused on pedestrian detection and tracking, mainly used in surveillance and night driver-assistance in cases of human are mostly walking upright. Actually, human’s motion manners are much more complex than description, so tracking human of arbitrary status is almost an impossible mission exactly. This paper is a part of the key research project of “Human Motion Object Recognition based on Infrared Images” founded by Ministry of Education of China (No.108174). In this project, the authors have some achievements on human segmentation and recognition in infrared images. The algorithm proposed in this paper is mainly to solve the tracking problem of human’s abnormal movements. The experiments results show that particles mean-shift migration algorithm proposed in this paper is independent on any complicated movement manner of human. Furthermore, viewing from the implementation process of the algorithm, it is very suitable for parallel computing.