改进GA-PSO优化SVM的行人检测算法

Pedestrian Detection Based on GA-PSO Optimized Support Vector Machine

针对当前对于行人检测的准确率和检测效率的要求越来越高,提出一种GA-PSO算法对于支持向量机(SVM)参数优化的行人检测方法。首先,针对梯度直方图特征描述子的维数高、提取速度慢,使用PCA对其进行降维处理;以SVM算法作为分类器,为避免传统单核支持向量机算法检测率低的情况出现,以组合核函数作为分类器核函数,并设置松弛变量,引进惩罚因子,结合遗传算法(GA)和改进权重系数的粒子群算法(PSO)进行组合系数和参数的优化与选择,根据优化后的参数构成最终的SVM分类器进行行人检测。实验结果表明,与传统SVM检测以及其他优化方法相比,检测率方面都有明显改进,且满足对检测效率的要求。

In view of the increasing demands for the accuracy and detection efficiency of pedestrian detection at present,a pedestrian detection method based on GA-PSO algorithm for the parameter optimization of support vector machine(SVM)is proposed.Firstly,aiming at high dimensionality and slow extraction speed of the gradient histogram feature descriptor,PCA is used to reduce the dimension.Secondly,SVM algorithm is used as a classifier,in order to avoid the low detection rate of the traditional single-core SVM algorithm,the combined kernel function is used as the classifier kernel functions,the slack variables are set,the penalty factors are introduced.Combining genetic algorithm(GA)and particle swarm optimization(PSO)with improved weight coefficient to optimize and select the combination coefficients and parameters,and the final SVM classifier for pedestrians is constructed according to the optimized parameters.The implementation results show that,compared with the traditional SVM detection algorithm and other optimization methods,the final pedestrian detection is better and meets the requirements of detection efficiency and accuracy.