基于方向梯度直方图和粒子采样定位的目标跟踪算法

Target Tracking Algorithm Based on HOG and Particle Sampling Location

针对静态背景和缓动背景下的多目标跟踪,提出一种基于HOG与粒子采样定位的单目标和多目标跟踪方法。从背景建模与更新策略出发,采用方向梯度HOG特征与朴素贝叶斯分类学习方法对检测的可疑目标进行判别,识别出兴趣目标。同时提出粒子采样定位算法,在初步确定的目标位置附近,利用一定分布特性的粒子对目标的位置状态进行逼近。对多场景多实例的跟踪仿真实验结果表明:该方法能够自动检测并判别兴趣目标,可以稳定跟踪单个或多个目标,并具有较高的定位精度。该方法可应用于静态背景和缓动背景下的单目标和多目标跟踪。

For the multi-target tracking under the static background or slowly moving background, the single target and multi-target tracking method based on HOG and particle sampling location is proposed in this paper. Starting from the back- ground modeling and updating strategies, the HOG features of the orientation gradient histogram and the NaYve Bayes classi- fication method are adopted to identify suspicious targets, and then the interested targets are identified. At the same time, the particle sampling location algorithm is proposed. In the neighbor of the initial target location, the location state of target is approximated by the particles of some distribution characteristic. The simulation experiments of multi-scene and multi-in- stance show that this method can detect and distinguish the interest target automatically, and can track a single or multiple targets stably, which has high precision. This method can be applied to the static background and slowly moving background of the single target and multi-target tracking.