Abstract: With the development of methods of gait recognition, it has become a hot direction of applying gait recognition system into real life. At the moment, the main obstacle hindering development of gait recognition system is the influence of changes of dress and carrying goods on gait profile of pedestrians. It proposes a gait recognition method based on track of gravity center, which firstly converses the original video of gait into binary image sequence and calculates the coordinates of gravity center in each frame of image, and then connects sequentially the coordinates of gravity centers of all frames in the image sequence to obtain the track of gravity center and multiplies it with scene factors to reduce the influence of viewing angle on coordinates of gravity center. The track of gravity center will be spectrum analyzed, and the spectrum transformed Fourier coefficients will be normalized to gait feature vectors and input into K cluster and BP neural network for recognition. The test proves that, the method is capable of achieving excellent recognition effect under conditions of different dresses and carrying goods and still maintains high recognition rate without retraining models when the pedestrians change their dresses and carrying goods.

Key words: gravity center track, scene factor, spectrum analysis, dress, carrying

摘要： 随着步态识别方法的发展，将步态识别应用到现实生活中已成为热点方向，目前影响步态识别系统开发的主要障碍是服饰和携带物的变化对行人步态轮廓的影响。提出了一种基于重心轨迹的步态识别方法，该方法首先将原始步态视频转化成二值的步态图像序列，对每帧图像计算重心坐标，将图像序列所有帧的重心坐标顺序连接得到重心轨迹，并乘以场景因子降低拍摄视角对重心坐标的影响。对重心轨迹进行频谱分析，将频谱变换的傅里叶系数归一化作为步态特征向量，输入神经网络中进行识别。实验结果证明该方法在不同种类的服饰和携带物条件下均具有很好的识别效果，当行人改变服饰、携带物时不需重新训练模型仍能保持较高的识别率。

关键词: 重心轨迹, 场景因子, 频谱分析, 服饰, 携带物