基于不变矩算法的行人识别方法研究

Pedestrian Recognition based on Invariant Moments Algorithm

行人识别对于智能辅助驾驶和智能车辆至关重要.采用一种基于不变矩算法的行人特征提取和识别方法,通过利用不变矩在目标平移、旋转和缩放的不变性,在HU不变矩基础上添加3个表达式,使不变矩包含更多的细节特征,将其作为行人目标的识别特征,利用支持向量机分类器作为主要手段对不变矩进行分类识别,并分析影响识别效果的影响因素.试验结果表明,选择改进的不变矩作为行人特征具有较好的行人识别效果,较高的识别率使行人和非行人能得到有效的识别.

Pedestrian recognition is essential for intelligent assisted driving and intelligent vehicles.A method of pedestrian feature extraction and recognition based on invariant moments algorithm is proposed.The invariant moments have the accustomed invariance for translation,rotation and scaling of targets which can be used as the invariant characteristic vectors.It includes more details of features which are added to three expressions on the basis of HU invariant moments.The support vector machine classifier is used to test the recognization of pedestrians,and the effect of factors affecting recognition is analysed.The experimental results show that the invariant moments can be selected as the pedestrian characteristics and can give good effectiveness for pedestrian recognition.The high rate of pedestrian recognition demonstrates that the proposed method can recognize the pedestrians and non-pedestrians effectively.