CAJ | 학술논문

为获得车载动态环境下较好的行人跟踪结果，提出一种融合梯度方向直方图（ HOG）与颜色直方图特征的粒子滤波行人跟踪方法。利用HOG特征训练识别行人的分类器获取待跟踪目标，利用颜色直方图粒子滤波预测行人位置状态信息，并在预测结果的一定范围内，利用HOG特征对行人进行验证及位置校正，不断修正粒子权值并实现重采样，实现动态背景下的行人跟踪。实验结果表明，与传统的粒子滤波算法相比，该算法能更准确有效地跟踪行人目标。
위획득차재동태배경하교호적행인근종결과，제출일충융합제도방향직방도（ HOG）여안색직방도특정적입자려파행인근종방법。이용HOG특정훈련식별행인적분류기획취대근종목표，이용안색직방도입자려파예측행인위치상태신식，병재예측결과적일정범위내，이용HOG특정대행인진행험증급위치교정，불단수정입자권치병실현중채양，실현동태배경하적행인근종。실험결과표명，여전통적입자려파산법상비，해산법능경준학유효지근종행인목표。
To achieve better pedestrian tracking results in dynamical vehicular circumstances , we present a pedestrian tracking method which is based on particle filter with integration of histogram of oriented gradients ( HOG) and colour histogram characteristics .HOG features are utilised to train the pedestrian recognition classifier for locating the targets to be tracked .Colour histogram particle filter is activated to predict the state information of pedestrians location , and within a certain range of the forecasted results , HOG features is utilised to verify the pedestrians and correct their locations .By constantly modifying the particle weights and re-sampling, the pedestrian tracking in dynamic background is realised.Experimental results show that the proposed algorithm can track pedestrian targets more accurately and effectively compared with traditional particle filter algorithm .