A Fast Non-parametric Detection Method for Infrared Small Dim Target

This paper presents a fast non-parametric detection method for multiple infrared small dim targets against the complex sky background. Traditional methods are based on the entire image processing with a high requirement of accuracy and speediness. However, the proposed method divides the detection into regions of interest (ROIs) searching and precise positioning according to the idea that from coarse searching to precise positioning. The former pays more attention to accuracy while the latter is more concentrated on speediness. First, ROIs sequence can be built by stationary wavelet transform (SWT) followed by Sobel direction directors. Through Hough line transformation, target region proposals which are settled as prior information of precise positioning can be extracted from ROIs sequence. Then, exact target positions are located by auto-regression analysis of temporal profile. Only the max size of targets is assumed which dramatically reduces the dependence of the method on target parameters. Results demonstrate that the proposed method can rapidly locate multiple targets simultaneously in real-time.