一种融合动态预测的感知哈希目标跟踪算法

为实现静态背景下的目标跟踪，采取基于DTC的感知哈希生成算法提取目标感知哈希摘要，比较汉明距离实现模板匹配。针对跟踪效率不足的问题，提出了一种范围与方位融合预测策略，其中范围预测是预测目标出现的范围，缩小模板搜索区域；方位预测在范围预测的基础上寻找一条便捷的搜索路线，将已跟踪得到的坐标要素简化为单一的方位角要素，结合预测模型预测目标方位，挖掘出一种模板匹配终止条件，在适当的深度终止窗口遍历。试验结果证明，范围预测对跟踪效率提升明显，预测范围缩小n倍时算法效率提升约n²倍，方位预测可在范围预测的基础上提升约16%的效率。

A perceptual Hash target tracking algorithm based on dynamic prediction

In order to achieve target tracking in static background, a DTC-based perceptual Hash generation algorithm is used to extract the target-aware Hash summary, and the Hamming distance is compared to achieve template matching. Aiming at the problem of insufficient tracking efficiency, a range and azimuth fusion prediction strategy is proposed, in which the range predicts the range of the predicted target and narrows the template search area. The azimuth prediction finds a faster and shorter search route based on the range prediction. The tracked coordinate elements are simplified into a single azimuth feature, the four prediction models are used to predict the target orientation, a template matching termination condition is mined, the window traversal is stopped at an appropriate depth. The experimental results show that the range prediction improves the tracking efficiency significantly. When the prediction range is reduced by n times, the algorithm efficiency is improved by about n² times, and the azimuth prediction can improve the efficiency by about 16% based on the range prediction.