CAJ | 학술논문

针对图像序列中多目标提取问题，提出一种向前估计跟踪的多目标检测算法。该算法首先通过单帧小目标检测算法，检测出图像中的疑似目标。然后，采用弹道滤波方法对连续帧中的疑似目标进行滤波。在弹道滤波的过程中，采用向后临近连续帧遍历组合的方法，形成新的假设弹道，并建立该弹道的向前多帧估计目标，避免了部分帧目标丢失，跟踪失败问题；采用高斯函数分配权值方法对同弹道同帧的估计目标进行有效融合，达到了目标估计偏差和目标圆度评价圆半径相当的水平，完成了实际目标弹道的建立。最后，通过弹道中目标点个数等状态信息，抛弃假弹道，实现多目标检测。该算法不需要人工导引，提高了多目标提取的自动化程度。
침대도상서렬중다목표제취문제，제출일충향전고계근종적다목표검측산법。해산법수선통과단정소목표검측산법，검측출도상중적의사목표。연후，채용탄도려파방법대련속정중적의사목표진행려파。재탄도려파적과정중，채용향후림근련속정편력조합적방법，형성신적가설탄도，병건립해탄도적향전다정고계목표，피면료부분정목표주실，근종실패문제；채용고사함수분배권치방법대동탄도동정적고계목표진행유효융합，체도료목표고계편차화목표원도평개원반경상당적수평，완성료실제목표탄도적건립。최후，통과탄도중목표점개수등상태신식，포기가탄도，실현다목표검측。해산법불수요인공도인，제고료다목표제취적자동화정도。
In order to extract multi-target from image sequence, a forward estimate tracking multi-target detection algorithm was proposed. First, the suspected targets were detected by a single frame small target detection algorithm. Then, the suspected target in the consecutive frames was filtered by ballistic filtering method. In the process of ballistic filtering, the new hypothesis trajectory was formed by using backward traversal close composition approach consecutive frames, and the multi-frame forward Estimation targets were established, which avoid the losing of multi-frame and tracking failures. By using Gaussian functions to assign weights to the same trajectory and same frame, the estimate targets were merged effectively. The projected estimated error is almost equal to radius of roundness evaluation. Finally, false trajectory was abandoned by using the number of ballistic targets and other status information, multi-target detection was achieved. The algorithm does not require manual guide, which improve the degree of multi-object extraction automation.