基于改进DLT算法的单目标识别跟踪研究

Research on Single Target Recognition and Tracking Based on Improved DLT Algorithm

为了提高DLT算法在视频中运动目标物体的跟踪鲁棒性、稳定性、实时性,对DLT算法进行改进。通过中值光流法计算初始位置和预测位置像素点灰度值的欧氏距离;运用一种新的在线学习机制P-N学习,用标记样本训练分类器,修改错误的正负样本,然后通过分类器对未被标注的数据进行标注,依次进行迭代训练;利用误差法和NCC法筛选出跟踪点,根据跟踪点的位置和距离的变化计算出目标框的大小和位置,实现目标跟踪。结果表明,改进后的DLT算法的目标跟踪的鲁棒性增强,能够根据目标框的实际情况获取有效的特征点,减少视频流中的目标过大或过小导致跟丢的情况。

In order to improve the robustness,stability,and real-time performance of the DLT algorithm in tracking moving target objects in videos,improvements have been made to the DLT algorithm.The Euclidean distance of the gray value of the pixels at the initial position and the predicted position is calculated with the median optical flow method.P-N learning method,a new online learning mechanism,is employed to train the classifier against labeled samples,correct incorrect positive and negative samples,and annotate unlabeled data.Such a process is repeated in iterative training.The error method and NCC method are used to filter out tracking points,calculate the size and position of the target box based on the changes in the position and distance of the tracking points to achieve target tracking.The results show that the improved DLT algorithm enhances the robustness of target tracking,captures effective feature points based on the actual situation of the target box,and reduces tracking failures due to a too large or too small target size in the video stream.