Pedestrian characteristics like bags, gender, clothes, or short hair might affect in identifying people in video surveillance. Due to variation in poses, illumination, background, and camera views, the fundamental problem in pedestrian attribute detection is the significant variation in visual manifestation and position of attributes. In this work, we use soft biometrics and appearance features to match people across cameras. The Convolutional Neural Network (CNN), the effective feature representation named Local Maximal Occurrence (LOMO), and the hier-archal Gaussian descriptor algorithm (GOG) are all being utilized to increase the performance of re-identification (re-ID). On the other hand, we applied soft biometric (SB) attributes to improve our system’s performance. The SB attributes are identified, and attribute-based distance scores are calculated between pairs of images (probe, gallery). The results based on soft biometric attributes can be used to define probe and gallery topics at a high level. We applied then the Random Forest (RF) clas-sifier as the primary classifier and a metric learning approach named Cross-view Quadratic Discriminant Analysis (XQDA) to compute the similarity between the probe and the gallery. The experimental results of our approach are evaluated on the VIPeR dataset. The results demonstrate that the SB approach gives promising potential and superior performance to its conventional counterpart methods (without soft).