Recognition of human targets in the scene belongs to the category of biometric identification, but compared with other biometric identification technologies such as face recognition, gesture recognition, fingerprint recognition, and iris recognition, infrared human target recognition is faced with unclear recognition categories, complex target structures, and Difficulty in positioning. Breaking through these constraints is the key to reliable identification of human templates. Based on the overview of the development status of thermal infrared human body target recognition technology, this chapter discusses the statistical classification and recognition methods of some thermal infrared human body targets.

 

Overview of Thermal Infrared Human Target Recognition

Since people are the main actors or participants in many real-world scenarios, the human body is an important work object for many thermal infrared imaging systems. For example, the vehicle thermal infrared night vision system identifies pedestrians when the vehicle is driving at night, warns the driver in time when a collision may occur, and plays a role in reducing and avoiding major personal injury accidents; the public area thermal infrared video surveillance system is in the monitored area. Identify and track human objects, analyze and judge whether there is inappropriate behavior, and alarm when necessary to achieve automatic security; thermal infrared gait recognition system recognizes human bodies on the basis of analyzing human gait and realizes identity authentication.

 

The difficulty of realizing correct and efficient infrared human target recognition mainly comes from the weak quality of thermal infrared images and the complexity of human objects, which are embodied in the following aspects.

(1) The human body is a non-rigid object, with various appearances and postures, and different shapes and sizes. In addition, the occlusion and adhesion between human objects and between human objects and other objects are also common. Because the human body's own characteristics do not have good stability and clusterability, in order to achieve human target recognition, the commonality of different human targets must be summarized, but the commonality is very unclear, and it is difficult to establish a reasonable model to describe it.

(2) Human body movement has strong randomness and randomness. Some joint parts of the human body can not only translate and rotate in three-dimensional space relative to the trunk part, but also have phenomena such as stretching, shrinking, bending, and occlusion. It is almost impossible to accurately locate these parts on the two-dimensional image plane. of.

(3) The brightness of the thermal infrared image target is not only related to the surface temperature of the target, but also to the surface characteristics, orientation, radiation wavelength and other factors of the target. For a human target, its brightness is also affected by these factors, and even if it is the same human target, for example, the thickness of clothing and different materials also affect its imaging brightness.

(4) Thermal infrared images have low contrast, poor layering, little texture information, and complete loss of color information, which makes human target recognition in thermal infrared images unable to use human skin color, clothing, It is more difficult to realize the fine features of eyes and faces, especially when there is occlusion and adhesion.

(5) The edges of human targets in thermal infrared images are blurred and have halo effects. In addition, there may be many interfering targets in the image, such as lamp posts, animals, vehicles, electrical boxes, buildings, etc. These heat source targets are easily confused with human targets.