一种迭代的核密度估计视觉目标检测算法

Iterative Visual Object Detection Algorithm Based on Kernel Density Estimation

利用srgb颜色空间性质,提出了一种基于Epanechnikov核函数的非参数核密度估计进行视觉目标检测,基于贝叶斯理论的迭代决策方法进一步消除噪声并增强了二值检测结果。在迭代过程中,模拟退火加快了收敛速度。仿真证明,算法能够抵抗光照变化和阴影等带来的不良影响,得到优异的目标检测结果;与几种经典方法相比,在大大降低误检率的同时提高了目标正确检出率,具有良好的实用价值。

Taking advantage of the property of srgb color space, a non-parametric kernel density estimation method based on Epanechnikov kernels was proposed to detect vision objects. The Bayes＇ Theory based recursive decision making technique further eliminated noises and enhanced the resultant binary image. In recursion, the simulated annealing boosted the convergence rate. Simulations prove the proposed algorithm can resist unfavorable effects like illumination changes and shadows, and can obtain an outstanding target detection result. Compared to several classical methods, it can dramatically reduce the false alarm rate meanwhile boosting the true detection rate, thereby has practical value.