CAJ | 학술논문

针对行人目标过小出现漏检和很难对具有多模态行人进行稳定跟踪等问题，提出一种高效和鲁棒的行人检测与跟踪方法。对于行人检测，设计了具有自适应调整滑动搜索窗口尺寸的梯度方向直方图(HOG)行人检测器搜索前景区域。在跟踪过程中，对每个行人建立基于前景区域的混合加权直方图模型，使用卡尔曼滤波对行人进行跟踪，获得候选区域并计算行人混合模型与候选搜索区域的匹配程度，同时根据匹配程度，给出一种混合模型在线更新策略，以确保跟踪的鲁棒性。实验结果表明，该方法提高了行人检测与跟踪的效率和准确率，解决了2人交叉导致跟踪失败的问题。
침대행인목표과소출현루검화흔난대구유다모태행인진행은정근종등문제，제출일충고효화로봉적행인검측여근종방법。대우행인검측，설계료구유자괄응조정활동수색창구척촌적제도방향직방도(HOG)행인검측기수색전경구역。재근종과정중，대매개행인건립기우전경구역적혼합가권직방도모형，사용잡이만려파대행인진행근종，획득후선구역병계산행인혼합모형여후선수색구역적필배정도，동시근거필배정도，급출일충혼합모형재선경신책략，이학보근종적로봉성。실험결과표명，해방법제고료행인검측여근종적효솔화준학솔，해결료2인교차도치근종실패적문제。
An efficient and robust method for pedestrian detection and tracking is proposed to solve the problems of leak de-tection for small objective and the difficulty in the stable tracking for multimodal pedestrians.Histogram of Oriented Gradi-ent(HOG)pedestrian detector with adaptive adjustment of sliding search window size is designed innovatively to search fore-ground area.In the tracking process,the mixed weighted histogram model based on foreground area is built for each person, the candidate areas are got by pedestrians tracking using Kalman filter,then the matching degree between pedestrians mixed model and candidate search area is calculated,meanwhile a hybrid model online update policy is made according to the matc-hing degree to guarantee the robust of tracking.Experimental results demonstrate that the method improves accuracy and efficiency of human detection and tracking,and solves the error tracking problem caused by the crossing pedestrians.