Person Recognition Based On Deep Learning Using Convolutional Neural Network
Minu Balakrishnan, G Dharanya, K Subaharini · 2022 IEEE 2nd International Conference on Mobile Networks and Wireless Communications (ICMNWC) · 2022
Matching the same subject across various cameras is the aim of the person recognition task. Earlier, person recognition problems were primarily addressed by image-based techniques. However, as the use of cameras and monitoring increases, image-based solutions are being employed. Image-based systems yield better results for person recognition because they take into account the person’s spatial and temporal information, which are not included in a single image. The CNN properties for person recognition are improved in this work. On huge annotated datasets like Image Net, recent studies have demonstrated the effectiveness of features produced from pre-trained Convolution Neural Network (CNN) top layers. In this study, we use a dataset of pedestrian attribute values to fine-tune the CNN features. We propose new labels that were produced by integrating a number of attribute labels included to the classification loss for the different pedestrian attribute labels. CNN is forced to develop more discriminative features that can identify more personally identifiable information as a result of combination attribute loss. Using traditional metric learning, we further enhance discriminative performance on a target identification dataset after extracting features from the trained CNN.