Dim small target tracking based on improved particle filter

With the rapid development of photoelectric technology, imaging detection system is widely used in navigation, astronomical target tracking and other fields. Dim small target tracking technology is one of the key research directions of imaging detection system. The performance of algorithm tracking directly affects the remote detection capability of the system.Following from two application backgrounds to introduce the difficulties faced by dim small target tracking of the imaging detection systems.1) aeronautical navigation is the earliest and most successful application field of imaging detection system. Because the distance between aircraft and imaging system is far away, the target only occupies several pixels on the imaging, and the radiation intensity of the target received by the imaging system is very weak, and it is easily disturbed by all kinds of noise and clutter.2) astronomical target orbit tracking is also an important application of imaging detection systems. In astronomical observations, due to the weak target intensity, it is basically submerged by noise and the signal-to-noise ratio is low. The time, position, target size and motion speed of the target are unknown, and no texture and other characteristics can be used. At the same time, because of the large amount of astronomical image data, the processing techniques such as image processing, data mining and signal processing are needed to deal with these massive data in time and effectively. In the above application fields, due to the existence of various objective factors, the target is easily overwhelmed by background clutter, so the quality of the dim small target tracking algorithm will directly determine the effective distance of the imaging system. For this reason, this paper considers that particle filter can solve nonlinear and non-Gaussian scenarios well, introduces it into the text, and improves it so that it can track the dim and small targets more stable.

    The research team of the Photoelectric Detection and Signal Processing Laboratory of the Institute of Optics and Electronics, Chinese Academy of Sciences is dedicated to the research of detection and tracking algorithms for photoelectric imaging targets.As to solve the problem of dim small target tracking in low SNR (SNR<3dB) scenes, an improved particle filter tracking method is proposed. This paper firstly obtains the gray feature by spatial position weighting method, and combines the neighborhood motion model and the gray probability graph to get the motion features of dim small target. Then construct the joint observation model of gray and motion features to calculate the particle weights. At the same time, in the process of tracking, the gray distribution of the target is not stable, and the strategy of adaptively updating the gray template of reference target is added. Finally, several groups of scenarios are used to verify the tracking effect of the proposed algorithm. Experimental results how that compared with the traditional algorithm, the proposed method greatly enhanced the tracking ability of dim small target in low SNR scenes. In the past two years, the latest research results of this research group include the automatic segmentation and analysis technology of biological living tissue images based on L1-L0, dim small targets detection based on self-adaptive caliber temporal-spatial filtering, dim and small targets detection based on local energy center of sequential image, dim small targets tracking based on the sample adaptive immune genetic particle filter algorithm, dim small target tracking based on improved particle filter, and so on. The corresponding contents are published in the international well-known journals, for example Biomedical Optics Express  and  Infrared physics and technology, and so on. Some representative tracking results are given below.

Tracking results. (a) The first frame; (b) The motion model graph; (c) tracking results

About Team
The profile of the Photoelectric Detection and Signal Processing Laboratory of the Institute of Optics and Electronics of Chinese Academy of Sciences are following. The main research directions of this group include image processing and analysis based on machine learning and deep learning technology, intelligent recognition, detection, high-precision tracking technology for various targets in complex scenes, and intelligent machines learning method based on the combination of neuroscience and information science, high speed parallel embedded information processing system structure design, real-time image processing and algorithm optimization, build an evaluation platform for target detection, recognition and tracking. The research members mainly include researchers with more than 30 professors, associate professors and more than 20 doctoral/master students. At present, the research team has participated in, undertaken and completed some researches including national 973, 863, ministerial and preparatory research projects in related fields. So far, more than 20 SCI/EI articles have been published in academic journals at home and abroad, and more than 60 academic papers have been published. Dozens of national invention patents have been applied, and some of them have been authorized.

Fan X S, Xu Z Y, Zhang J L. Dim small target tracking based on improved particle filter[J]. Opto-Electronic Engineering, 2018, 45(8): 170569.