基于改进多尺度特征估计的行人检测算法

Pedestrian Detection Algorithm Based on Improved Multi-scale Feature Approximated

多尺度特征在行人检测中广泛应用,但计算效率较低,而多尺度特征估计算法不仅能够确保行人检测效率,还能使行人检测精准度达到较好的水平。文中提出了一种改进的多尺度特征估计方案,根据自然图像像素的对比度具有尺度不变性的特点,对部分样本在较小的尺度空间内用最小二乘拟合方法拟合出估计函数,用于其它样本多尺度特征的估计。实验结果表明,该方案能够在保证行人检测效率的前提下,进一步的提高行人检测精准度。

Multi-scale feature is widely used in pedestrian detection,but the efficiency is lower.The algorithm of multiscale feature approximated can guarantee the efficiency and the accuracy.An improved method of approximating multi-scale feature is presented,according to the scale-invariant of the pixel＇s contrast of natural image,the least squares fit is used to approximate the function in small range of scales for sub-sample,which is used to approximate multi-scale feature of other samples.The experiment results show that this method can improve the accuracy of pedestrian detection without sacrificing detection efficiency.