Infrared Small Dim Multi-targets Real-time Processing

The infrared imaging technology has strong concealment and good anti-interference ability,which makes it widely used in civil and military fields.Because of remote imaging and background clutter interference,signal-to-noise ratio of the infrared image is low and the structure information is so insufficient,the research of infrared image preprocessing method,small dim target detection and small dim target tracking methods become the key of infrared small dim multi-target real-time processing technology,and in some important field such as infrared guidance,it has great significance.In this paper,it realized the design and optimization of the infrared image real-time processing method based on FPGA+DSP architecture from the practical research project.The improved median filter was used to adapt to different noise types and noise densities.The window shape was changed while changing the length of the sliding window with smaller windows.Based on the target and background characteristics,the background suppression algorithm is used to extract the background of undulating changes by the half-circle combined structure element based on changed scale,and the effective suppression of background clutter was realized.In the case of small dim target detection,the point spread model with scale factor was used to represent the image under Laplace Gaussian space,and the target position and size were determined initially.All the suspicious objects were extracted by the threshold of the mean difference,and then according to the target size combined with different degrees of differentiation to achieve the extraction of real goals.In the case of weak multi-target tracking,in order to accommodate different maneuvering target tracking,the low-speed moving target matched the mean drift of the Kalman filter while the high maneuvering target matching improved mean shift particle filter in order to achieve reliable tracking of multi-objective,the state of the adjacent target of each target was considered by combining the Markov stochastic network to estimate the maximum combined posterior probability of each target,and update the filter parameters and particle weights,finally,it could realize multi-target location estimation,Based on the infrared real-time image processing platform,the large amount of continuous image data from infrared wave camera with 640*512 resolution ratio were processed by the algorithm.The results show that the improved median filter noise smoothing algorithm and the improved morphological background had good processing effect and with real-time performance.In the case of multi-target detection,the scale-based detection algorithm has better robustness,higher detection rate,and the average processing time of single frame was less than 5s,so it met the real-time requirement.In the case of multi-target tracking,the multi-model improved Kalman particle filter based on Markov stochastic network is better than the traditional interaction Multi-model algorithm,its processing speed can reach 72 frames / s,and with high reliability.In conclusion,the infrared weak multi-objective real-time processing method is reliable and has high practical application value.