Person Re-Identification using Deep Learning

Person re-identification is a task which is used to search the particular person captured in one camera vision against a set of images captured in other camera visions. When aprobeimageisgiven,theperson re-identification model has to find theclosestmatches to the probe image from gallery images and arrange the gallery images according to the similarity score. This task has challenges because of variations in lighting, pose, background clutter, scale, occlusions and similar clothing. The primary goal of this research work is to enhance the performance of the person re-identification model in terms of identification rate at RankI and mean Average Precision (mAP). To achieve this goal and overcome the limitations, the research objectives were framed to develop an unsupervised person re-identification model to overcome the problem of limited training samples and reduce computation complexity; identify an enhanced feature engineering methodology that results in reduced feature vector size and elimination of unwanted background information from the given image; design and develop a deep neural network model to effectively perform with lesser training samples and also incorporate the spatial relationship between the features; test the models and evaluate the performance using benchmark datasets. In this research work, deep learning model with ResNet50 architecture has been proposed to increase the identification rate.