在面对光照变化、部分遮挡、背景杂乱和平面内外旋转等跟踪难点时, 跟踪学习检测算法(Tracking-Learning-Detection, TLD)容易产生漂移导致跟踪失败, 其跟踪性能还有待提高。在传统TLD算法的基础上, 提出一种基于人工鱼群粒子滤波的TLD改进算法。首先使用人工鱼群粒子滤波跟踪器代替金字塔光流跟踪器, 将颜色直方图特征和方向梯度直方图特征进行融合, 建立目标表观模型, 引入图像金字塔多尺度思想进行尺度匹配, 提高目标跟踪的稳健性。然后通过粒子滤波过程预测目标区域, 将TLD算法检测模块的全局扫描改进为局部扫描, 剔除大量非目标区域, 提高检测模块的检测效率。实验结果表明: 基于人工鱼群粒子滤波的TLD改进算法具有良好的跟踪性能, 与传统TLD算法相比, 其平均成功率和精准度分别提高了19.04%和28.00%, 平均跟踪速度可达33.87 FPS, 提高了38.78%。

Faced with tracking difficulties such as illumination variation, occlusion, background clutters, in-plane and out-of-plane rotation, the Tracking-Learning-Detection (TLD) algorithm is prone to drift and cause tracking failures, and its tracking performance needs to be improved. Based on the traditional TLD algorithm, an improved TLD algorithm based on artificial fish swarm particle filtering is proposed. Firstly, the artificial fish swarm particle filter tracker is used to replace the optical flow tracker. The color histogram feature and histogram of oriented gradient feature are fused to establish the target appearance model. The image pyramid multi-scale idea is introduced for scale matching to improve the robustness of target tracking. Then, the target area is predicted through the particle filtering process. The global scanning of TLD algorithm detection module is improved to be the local scanning. A large number of non-target areas are eliminated, and the detection efficiency is improved. The experimental results show that the improved TLD algorithm based on artificial fish swarm particle filtering has good tracking performance. Compared with the traditional TLD algorithm, its average success rate and precision have improved by 19.04% and 28.00%, and the average tracking speed can reach 33.87 FPS, which has improved by 38.78%.