针对在复杂环境中目标尺度变化、形状变化以及场景光照变化、背景干扰等因素导致的目标跟踪稳定性下降问题,提出一种基于自适应多层卷积特征决策融合的目标跟踪算法。首先,通过卷积神经网络VGG-Net-19提取目标候选区域的多层卷积特征;其次,在相关滤波模型框架下,利用这些卷积特征构建多个弱跟踪器;接着,根据每个弱跟踪器的决策损失变化自适应地调节它们的决策权重,完成基于多层卷积特征的目标位置估计;然后,根据尺度相关滤波模型在目标中心区域进行多尺度采样,并利用相邻帧的尺度变化先验分布完成对目标尺度的预测。选取51组具有多种挑战因素的视频序列对所提算法的跟踪性能进行测试。实验结果表明,与当前主流的目标跟踪算法相比,所提算法取得了更高的跟踪精度和成功率,同时可以较好地适应目标的尺度变化,并且在目标发生形变、场景出现光照变化及背景干扰等复杂条件下仍具有较好的跟踪鲁棒性。

To address the tracking stability degradation caused by target scale variation, deformation, illumination variation, and background clutter in complex scenes, a target tracking algorithm based on adaptive multilayer convolutional feature decision fusion is proposed. Initially, multilayer convolutional features are extracted from a target candidate region using the VGG-Net-19 convolutional neural network. Then, under a correlation filter model framework, the extracted convolutional features are employed to construct several weak trackers. Decision weights are adjusted adaptively based on the fluctuation of the decision losses of these weak trackers, and the target position is estimated based on the multilayer convolutional features. Next, according to a scale correlation filter model, multiple scale image patches are sampled at the target center position. Taking advantage of the prior distribution of scale variation between adjacent frames, its scale is predicted. Fifty-one video sequences with multiple challenging attributes are selected to evaluate the tracking performance of the proposed algorithm. The experimental results demonstrate that the proposed algorithm has higher tracking accuracy and success rate compared with state-of-the-art target tracking algorithms. The proposed algorithm adapts well to target scale variation. In addition, it improves the target tracking robustness under target deformation, illumination variation, and background clutter conditions.