CAJ | 학술논문

CAMShift目标跟踪算法遇到目标被遮挡时容易陷入局部最大值，对快速运动目标容易跟踪失败，且无法从失败中复原。针对该问题，利用Kalman预测器改进CAMShift算法。首先利用Kalman预测器预测下帧图像中目标的位置，以此位置为中心确定 CAMShift 算法进行目标跟踪的搜索区域；然后利用目标匹配时的Bhattacharyya系数及目标大小来判断目标是否被遮挡以及被遮挡的程度。如果没有被遮挡，则用CAMShift算法得到的目标位置更新Kalman预测器中参数；如果遮挡不严重，则用Kalman预测器的预测值作为目标的位置和大小，且用该组值更新Kalman预测器中参数；如果遮挡非常严重，则用Kalman预测器的预测值作为目标当前位置，目标大小为固定值，用该组值更新Kalman预测器中参数。实验结果表明，改进算法能够准确地跟踪被遮挡目标和快速运动目标。
CAMShift목표근종산법우도목표피차당시용역함입국부최대치，대쾌속운동목표용역근종실패，차무법종실패중복원。침대해문제，이용Kalman예측기개진CAMShift산법。수선이용Kalman예측기예측하정도상중목표적위치，이차위치위중심학정 CAMShift 산법진행목표근종적수색구역；연후이용목표필배시적Bhattacharyya계수급목표대소래판단목표시부피차당이급피차당적정도。여과몰유피차당，칙용CAMShift산법득도적목표위치경신Kalman예측기중삼수；여과차당불엄중，칙용Kalman예측기적예측치작위목표적위치화대소，차용해조치경신Kalman예측기중삼수；여과차당비상엄중，칙용Kalman예측기적예측치작위목표당전위치，목표대소위고정치，용해조치경신Kalman예측기중삼수。실험결과표명，개진산법능구준학지근종피차당목표화쾌속운동목표。
CAMShift(Continuously Adaptive Mean Shift) target tracking algorithm is liable to fall into a local maxima when the target is occluded, and is prone to failure when the targets move fast, and can not be recovered from the failure. To solve this problem, the CAMShif algorithm is improved by using Kalman predictor. First, the position of the target in the next frame image is predicted by using the Kalman predictor and this position is used as the center to determine the searching area of CAMShift target tracking algorithm. Then the Bhattacharyya coefficient of target matching and the size of the target are utilized to determine whether the target is occluded and the degree of occlusion. If not occluded, the parameters of Kalman predictor are updated by the position of the target with CAMShift algorithm. If the occlusion is not serious, the current location and size of the target are determined by the predictive values of Kalman predictor, and this set of values are used to update the parameters of Kalman predictor. If the occlusion is very serious, the current location is determined by the predictive values of the Kalman predictor and the target size is a fixed value, then this set of values are used to update the parameters of Kalman predictor. The experimental results show that the improved algorithm is able to accurately track the occluded and/or fast moving targets.