Study On Technology Of Feature Extraction And Recognition Of Dynamic Target

In the system of target identification and tracking, the tracking point woulddeviate or be lost due to the changing of size, distance, pose or the background of theaircraft. The tracking would even fail. So it is extremely important to find a stabletracking point. The tail fire is the clearest, least interferential and easiest part torecognize within an infrared image. We found the head of tail fire, we find the aircraft,and we find the stable point for the tracking of aircraft. So the target of this paper is toidentify the head of tail fire by using the morphological characteristics.This paper has three parts: segmentation, feature extraction and targetidentification.The aim of segmentation is to separate the tail fire completely from backgroundand the aircraft. The results would affect feature extraction and recognition directly.Since the brightness value of tail fire is much higher than the background and aircraft,this paper choose threshold method to segment. This paper presents a segmentationalgorithm based on area evaluation of binary image. First segment the image by Otsu.Then calculate the area ratio of target and compare it to0.5%. If the ratio is less than0.5%, re-segment the original image by the improved Otsu algorithm. This algorithmand three common thresholding algorithms are simulated and compared. The resultsshow that the algorithm can adaptively segment the tail fire according to its size.Chain-code is chose to extract the contour of the target. The traditional method togain the chain-code is to traverse the contour of the binary image. But during theengineering application, the image is saved as segment-form to reduce memory space.The existing algorithm of converting segment-form to chain-code form is incomplete.The program would blow up when the target is special. So this paper proposes animproved converting algorithm based on the existing one. Experimental results showthat the new method achieves the desired aims.Finally, the head region of tail fire is extracted from the chain-code table based onthe Chain-code-sum and the Chain-code-difference. Combine the principal axes, theprecise position of head is found. Experimental results show that this algorithm has ahigh accuracy.