In non-overlapping multi-camera surveillance systems person re-identification is one of the main issues. Aiming for person re-identification useing statistical properties of the objects and features by training

we propose a method by combining global and local features to identify the same person in different images. This method does not need a training phase

and it is robust to different viewpoints

illumination changes

and varying poses. First

the object is recognized roughly by spatiograms. Then the human target is divided into three parts. By ignoring the head part

the local color and shape features of the main body

the arms and the legs are extracted. Thus

the recognition of the person is carried out according to the Earth movers distance of the local features. The experimental results show that the proposed method has a higher accuracy rate

and it is invariant to the effects of occlusion and background adhesion.