System for tracking infrared human body target based on corpuscle dynamic sampling model

Abstract

The invention provides an infrared human target tracking system based on a particle dynamic sampling model. The system includes a thermal infrared imager, an image collection device, a CPU and a video display device. The image collection device collects and transmits the video data flow of the thermal infrared imager to the CPU; the CPU combines the infrared video data into a continuous image sequence and carries out tracking on the image sequence. A tracking method is realized by random particle sampling and transfer control, which includes: taking an individual target as an object, regenerating a random sampling particle according to the observing particles of each target and removing the redundant particles according to the set judging rule of an ineffective sampling particle; carrying out state judging on the targets in a tracking pool; the tracking result processed by the CPU is fed into the video display device to carry out display; the technology facing to the object is adopted to carry out sealing on the complex target states of the human; the problem that the traditional tracking method can not realize the reliable tracking to the human target in the infrared image sequence is overcome.

Description

technical field [0001] The invention relates to a tracking system of a human body target, in particular to an infrared human body target tracking technology based on a particle dynamic sampling model and an object-oriented method. Background technique [0002] Compared with visible light imaging, the advantages of infrared imaging are self-evident: no auxiliary light source is required, and it can work around the clock in almost any weather conditions and environments. With the development of infrared device manufacturing technology, especially the uncooled focal plane infrared CCD manufacturing technology and the decline in price, the application field of infrared imaging system has been rapidly expanded. [0003] The human target is the most active and valuable element in the scene, so the application of tracking technology for complex non-rigid targets such as human targets in infrared images is very wide. The main application fields include: infrared aiming system, infr...

AI Technical Summary

Problems solved by technology

[0008] Due to the high subjective randomness of the human target, the target may change the speed, direction, form and path of the movement at any time when the observer cannot predict it. That is, in most cases, the movement of the human body has no rules to follow, and the movement model with a fixed dynamic system cannot be used to describe the movement laws of the human body

This makes some tracking methods based on motion law assumptions, such as the Kalman filter method, unable to effectively track human targets.

[0009] (3) Defects of infrared imaging itself

[0010] Since the infrared image is a grayscale image without the color information of the target, the tracking method based on color features such as skin color and clothing, which is widely used in visible light images, cannot be used; due to the lack of texture details of the human target in the infrared image, the local texture expression of the target does not have Good identifiability, some non-rigid object tracking methods based on texture similarity calculation in the visible light image target tracking method, such as the traditional Mean Shift tracking method, conduct similarity analysis through the Bhattacharyya coefficients of the front and rear two regions, and maximize the Bhattacharyya coefficient The tracking window area is used as the matching area, which is not applicable in the infrared human target tracking process, because the gray histogram features of different targets are very close, and the feature quantity used to judge the difference between different targets is relatively small. Less, using the Bhattacharyya coefficient of the gray histogram for similar matching has a high possibility of mismatching

[0011] (4) Independence of human target detection algorithm

[0012] In addition to the unreliability of the detection algorithm caused by the complex changes in the posture of the human target, the image of the human target is often incomplete and broken due to the interference of the background, and has completely lost the basic characteristics of the human target. Even if the detection algorithm is 100% Reliable, human targets cannot be identified at this time, so a tracking solution based entirely on detection algorithms is not feasible

[0013] The tracking of infrared human targets at home and abroad is based on the assumption of the state of the target: such as assuming that the target is moving at a constant speed, assuming the posture of the target is an upright state (pedestrian), assuming that the camera is stationary, etc., using Kalman filter, Particle filter or Mean Shift method for tracking based on state prediction method and similarity analysis, the application occasions are very limited and the tracking results are not ideal

Images