基于上下文敏感度的动态图像目标识别

Dynamic Image Target Recognition Based on Context Sensitivity

针对机载实时系统获取的图像序列既不满足背景固定又没有目标、背景的高灰度比,而且常规方法处理这些图像存在局限性,提出了基于上下文敏感度的运动图像目标识别算法.首先,对标准Hough变换(HT)算法进行改进,即沿着某一方向做HT,如果线段长度或已识别的目标点数加上剩余线段长度小于阈值,则停止此方向的扫描,这种变换克服了标准HT以图像边界点为扫描边界的缺点.其次,在静态图像识别的基础上,定义了图像识别的置信度,对运动图像前后帧的关系进行了量化,据此实现了对目标的识别.实验结果表明,与标准HT相比,改进的算法不仅具有很强的稳定性和高抗干扰能力,识别准确性高,而且处理速度是标准HT的3～5倍.

Since the image sequences acquired from the on-board real-time system can not satisfy fixed background, neither have target and high grey ratio of the background, so that there exist defects by using traditional methods to deal with these images. A moving target tracking method based on context sensitivity was put forward. Firstly, an improved Hough transform (HT) algorithm was proposed to recognize the target from static images, in which if the length of the line, or the recognized number of target points plus the remnant length of the line are less than the threshold along certain direction, then the scanning is stopped in that direction, so the image edges are not used as scanning edges as done in the standard Hough transform. Secondly, based on the static image recognition algorithm, the believe degree of the recognized image is defined and the contextual relation of the image sequence is quantified, and then the target recognition is realized. Experiment results show that, compared with the standard Hough transform, the proposed method not only has stronger stability and anti-interference ability, but also has higher recognition accuracy, and the processing speed is 3-5 times of the standard Hough transform.