CAJ | 학술논문

针对传统的 Mean-Shift 跟踪算法，使用单个颜色特征定位目标易受相似目标与背景的干扰导致跟踪失败以及跟踪窗口尺寸不能自适应跟踪目标变化的问题，提出一种基于颜色特征与边界特征相融合的目标表示方法和沙包核函数 Mean-Shift 尺寸自适应算法。在跟踪中，颜色特征和边界特征根据各个特征的可靠性进行实时性更新；同时，在跟踪窗口中心和边界定位的基础上，由候选目标跟踪窗和分块目标跟踪窗的边界距离变化对核窗宽大小进行更新。实验结果表明，该算法目标定位的精确性更高，在目标尺寸增大和减小的情况下，平均每帧耗时比传统的基于矩形窗和椭圆窗自适应跟踪算法更少，提高了跟踪性能，满足实时性要求。
침대전통적 Mean-Shift 근종산법，사용단개안색특정정위목표역수상사목표여배경적간우도치근종실패이급근종창구척촌불능자괄응근종목표변화적문제，제출일충기우안색특정여변계특정상융합적목표표시방법화사포핵함수 Mean-Shift 척촌자괄응산법。재근종중，안색특정화변계특정근거각개특정적가고성진행실시성경신；동시，재근종창구중심화변계정위적기출상，유후선목표근종창화분괴목표근종창적변계거리변화대핵창관대소진행경신。실험결과표명，해산법목표정위적정학성경고，재목표척촌증대화감소적정황하，평균매정모시비전통적기우구형창화타원창자괄응근종산법경소，제고료근종성능，만족실시성요구。
The classical Mean-Shift algorithm is only used single color histogram represents the target which easily leads to track failure,especially if the scene contains other objects characterized by a color distribution similar to that of the object of interest.The other hand,window function can not be adaptive in the object tracking.So the paper proposed a new algorithm for tracking target that combined color histogram with boudary describing goal,wihch was the real-time update according to the re-liability characteristics.As well as it proposed a newly adaptive window of sandbags which was adaptive based on the size change between the candidate and the block target.Take the experimental result into consideration,the paper proposed new al-gorithm’s had well accuracy in target location.The new algorithm’s average per frame time-consuming is more decreased than the traditional adaptive tracking algorithm based on a rectangular window and elliptic window when the goal scale getts bigger or smaller.It improves the tracking performance,meets the real-timer equirements.