基于SUSAN角点和HSV颜色模型的粒子滤波目标跟踪算法

PARTICLE FILTER OBJECT TRACKING ALGORITHM BASED ON SUSAN CORNER DETECTION AND HSV COLOUR MODEL

针对传统粒子滤波目标跟踪算法在目标与背景颜色相似情况下目标定位偏差大、易导致丢失目标的缺陷,提出一种基于角点和颜色模型的粒子滤波目标跟踪算法。首先,提出一种改进SUSAN角点检测算法,采用圆形模板邻域内像素灰度值中值代替模板中心像素灰度值作为模板"核"来检测区域目标角点,其改进SUSAN角点算法在继承原有SUSAN算法计算简单、定位准确、具有旋转不变性等特点的同时,具有更好抗噪声性能;其次,利用HSV颜色模型光照不敏锐特性,对检测到的角点建立HSV颜色模型,并将其嵌入到粒子滤波框架中,实现对目标的跟踪。实验结果表明,当背景与目标颜色相近时,该算法能够有效避免背景对目标的干扰,取得了较好的目标跟踪性能。

For deficiencies of traditional particle filter object tracking algorithm in bigger deviation of object localisation and being prone to object missing when its colour is similar to background,we proposed a particle filter object tracking algorithm which is based on corner and colour model. First,we presented an improved SUSAN corner detection algorithm,which uses median grayscale value of the pixel in neighbourhood of circular template instead of the greyscale value of the pixel in template centre as the ＂kernel＂of the template to detect regional object corner,the improved SUSAN corner algorithm not only inherits the characteristics of original SUSAN algorithm,such as simple calculation,accurate positioning, rotation invariance, etc., but also has better anti-noise performance. Secondly, we utilised the characteristic of illumination invariant in HSV colour model to build HSV colour model for the detected corner,and then embedded it into the particle filter framework to realise object tracking. Experimental results showed that when the background closed to the target colour,this algorithm could effectively prevent the interference of background and achieved a better object tracking performance.