复杂场景下基于自适应特征融合的目标跟踪算法

为提升复杂场景下目标跟踪的鲁棒性，优化模型运行效率，提出一种基于自适应特征融合的相关滤波跟踪算法。该算法采用方向梯度直方图特征和卷积神经网络来对目标进行信息构建，利用特征响应的峰值旁瓣比和旁瓣值占比自适应地确定融合系数，根据融合响应来预测目标位置。为适应场景的变化，降低光照、背景和目标形变等对跟踪的影响，引入平均峰值相关能量来设计滤波器学习率调整机制，动态地进行模型更新。通过对深度特征提取网络进行轻量化设计，降低特征网络参数，提高跟踪速度。在OTB100通用数据集上进行测试，实验结果表明:文中所提算法有效降低了干扰对目标跟踪的影响，且跟踪精度、成功率和速度整体优于对比算法。

Target tracking algorithm based on adaptive feature fusion in complex scenes

In order to improve the robustness of the target tracking algorithm in complex scenes and optimize the operating efficiency of the model, a correlation filter tracking algorithm based on adaptive feature fusion was proposed. The algorithm adopts histogram of oriented gradient feature and deep feature extraction network feature to construct the target information, uses the peak to sidelobe ratio and the value of side lobe ratio of feature response to adaptively determine the fusion coefficient, and predicts the target position according to the fusion response. In order to reduce the influence of illumination variation, occlusion and target deformation on the tracking process and adapt to the change of scene, the average peak-to correlation energy was introduced to design the filter learning rate adjustment mechanism and update the model dynamically. Through the lightweight design of the deep feature extraction network, the parameters of the feature network were reduced and the tracking speed was improved. Experimental results show that the algorithm effectively reduces the influence of interference on the tracking results, and the algorithm has better performance in tracking precision, success rate and speed compared with other tracking algorithm on the public video dataset OTB100.