基于特征融合和尺度自适应的干扰感知目标跟踪

发布时间：2018-06-20 22:57

本文选题：机器视觉 + 目标跟踪　；参考：《光学学报》2017年05期

【摘要】：针对复杂场景下单一颜色特征稳健性差、存在类目标干扰及目标尺度变化的问题,提出了一种基于特征融合和尺度自适应的干扰感知目标跟踪方法。首先,综合目标、邻域背景、类似干扰区域的三原色(RGB)特征和改进的方向梯度直方图(HOG)特征计算得到干扰感知目标模型;在搜索区域内逐像素点计算目标概率图,然后进行密集采样得到候选目标,利用目标概率图的概率值与距离值进行加权,同时定位目标和类似干扰,并更新目标模型;采用RGB直方图建立尺度模型,从当前帧图像上截取不同尺度的图像块并计算其RGB直方图,通过与尺度模型比较,获得最优尺度估计并更新尺度模型。实验结果表明,提出的方法对复杂场景下的类目标干扰、局部遮挡、尺度变化等均具有很好的适应性,同时距离精度、重叠精度等指标优于对比算法。
[Abstract]:In view of the poor robustness of single color feature in complex scene, the problem of class target interference and target scale change, a target tracking method based on feature fusion and scale adaptive is proposed. Firstly, integrated target, neighborhood background, similar RGB features and improved direction gradient histogram (HO) G) the target model is calculated by the feature calculation, the target probability map is calculated in pixel point by pixel in the search area, then the candidate target is obtained by dense sampling, the probability value of the target probability map is weighted with the distance value, the target and the similar interference are located, and the target model is updated. The scale model is established by using the RGB histogram, from when the scale model is established. In the front frame, the image blocks of different scales are intercepted and their RGB histogram is calculated. By comparing with the scale model, the optimal scale estimation is obtained and the scale model is updated. The experimental results show that the proposed method has good adaptability to the class target interference, the local occlusion, the scale change in the complex scene, and the distance precision and the overlap essence. The index of degree is better than the contrast algorithm.
【作者单位】：南京理工大学自动化学院;
【基金】：江苏省博士后科研自助计划(1601181B)
【分类号】：TP391.41

【相似文献】