中文摘要：

针对传统的基于梯度方向直方图（HOG）特征的行人检测耗时较长的问题，提出了基于腿部HOG特征优化的行人检测方法。该方法采用加权Fisher线性判别（WLFD）代替线性SVM来选择最具区分性的HOG特征，在保持分类能力的同时减少训练时间和存储空间，而且选择查找表型弱分类器的GentleAdaboost算法来训练优化权重组合HOG特征，形成一个强分类器来检测行人。通过对线性SVM、加权Fisher与阈值型以及加权Fisher与查找表型三种弱分类器的对比试验表明，基于加权Fisher与查找表型HOG特征优化后不仅提高了检测精度，而且训练时间和检测时间也能明显降低。

英文摘要：

To solve the time consuming problem of traditional pedestrian detection methods based on histograms of orien- ted gradients （HOG） features, a novel pedestrian detection method based on optimization of the HOG features of legs is presented. The method uses weighted linear Fisher discriminant （WLFD） instead of linear support vector machine （SVM） to construct week classifiers with the aim of selecting high discriminative HOG features, which can significantly decrease the training time and memory while maintaining the comparable classification performance. Moreover, the look up table （LUT） Gentle Adaboost algorithm is selected to optimize the weighted combined HOG features and form a strong classifier to identify the pedestrian. The comparison test shows that the classifier of WLFD with LUT outperforms the weak classifiers of the linear SVM and the WLFD with stump. When the HOG fea- tures are optimized by the classifier of WLFD with LUT a higher detection accuracy with lower training and detection time can be achieved.