CAJ | 학술논문

传统的多尺度特征计算都是首先构造不同尺度的图像形成图像金字塔，然后在金字塔每一层上通过滑动窗口的办法提取相应的特征，实验表明在目标检测时特征提取消耗大量时间，改善特征提取的速度是提升目标检测速度的关键。本文使用FFC(Fast Feature Computation)计算方法对多尺度图像特征进行快速提取，同时结合Adaboost算法和多特征融合方法用于行人目标检测，实验结果表明效果较好。
전통적다척도특정계산도시수선구조불동척도적도상형성도상금자탑，연후재금자탑매일층상통과활동창구적판법제취상응적특정，실험표명재목표검측시특정제취소모대량시간，개선특정제취적속도시제승목표검측속도적관건。본문사용FFC(Fast Feature Computation)계산방법대다척도도상특정진행쾌속제취，동시결합Adaboost산법화다특정융합방법용우행인목표검측，실험결과표명효과교호。
Traditional multi-scale feature extraction method for pedestrian detection firstly constructs image pyramid and then extracts corresponding features by sliding window method. The experiment reflects that it costs much time on feature extraction while we detect pedestrians. What we have to do is improving feature computation speed while does not matter detection perfor-mance. This paper proposes FFC(Fast Feature Computation) method for feature extraction and makes use of Adaboost algorithm and multi-features merge. Experimental results show that the effect is better.