模板匹配跟踪的哈希增强算法

HASH ENHANCEMENT ALGORITHM FOR TEMPLATE MATCHING TRACKING

基于平方差匹配SSD(Sum of Squared Difference)、标准相关系数匹配NCC(Normalized Cross Correlation)的传统模板匹配跟踪算法十分依赖灰度信息。图像灰度剧烈变化或光照强度改变明显的情况下容易被图像的高频信息影响,导致产生目标偏移的现象,跟踪精度降低。提出一种模板匹配跟踪的哈希增强算法,用于减小灰度与亮度变化的影响,增强目标跟踪的精度。其具体过程是:以感知哈希提取感知特征,对提取的特征二值化生成跟踪目标的哈希序列;将哈希序列作为匹配跟踪的模板,通过比较汉明距离在每一帧中寻找最为相似的目标以达到跟踪的效果;运用抽屉原理缩短汉明距离比较的时间。给出一种基于汉明距离的自适应模板更新算法,保证了跟踪的连续性。仿真实验结果与基于SSD、NCC的模板跟踪算法比较,在目标灰度或亮度变化情况下匹配精度高出一个数量级,跟踪速度能够满足实时性需求。

Traditional template matching tracking algorithms based on sum of squared difference( SSD) and normalised cross-correlation( NCC) depend on grey information very much. In circumstance of dramatic greyscale change or significant illumination intensity variation in image,they are prone to be affected by high-frequency information of the image and this leads to the phenomenon of target drifting followed by the decrease in tracking accuracy. We proposed a hash enhancement algorithm for template matching tracking used to reduce the impacts of greyscale and luminance variations,and to enhance the tracking accuracy as well. The specific progress is as follows: It extracts the perceptual features of the target with perceptual hashing,and binarises the extracted features to generate hash string of the tracking targets.Then it saves the hash string as a template of matching tracking,by comparing the hamming distances it searches the most similar targets from each frame so as to reach the effect of tracking. By using drawer principle it shortens the time of hamming distances comparison. We also proposed a hamming distance-based adaptive model updating algorithm which guarantees the continuity of tracking. Comparing the results of simulative experiments with SSD-based and NCC-based template tracking algorithms,the matching accuracy reaches an order of magnitude higher in the case of target's greyscale and luminance varying. What's more,the tracking speed is able to meet the requirement of real-time property.