Facial and Body features based Multi-Model Person Re-Identification: MPRe-ID

Within the surveillance area, Person Re-identification (Re-ID) holds considerable importance by enabling the matching of a person's appearance across multiple non-overlapping cameras.Nonetheless, this task poses challenges due to factors like changes in camera viewpoints, occlusion, and variations in appearance, including clothing, shoes, and pose.Overcoming these challenges requires discriminative feature learning.Deep convolutional neural networks (CNNs) have recently gained widespread usage to address this objective.This study introduces a lightweight and robust deep learning framework Multi-Model person Re-ID (MPRe-ID) for person re-identification.It incorporates the YOLOv4 object detection model for pedestrian detection and utilizes SORT with deep metric association (DeepSORT) algorithm for tracking.MPRe-ID uses novel body feature extraction model to learn discriminative features at various semantic levels, leveraging the ResNeXt architecture as its backbone.The proposed body feature extraction model contains multiple blocks where channels are concatenated between blocks, and an aggregation gate is employed to aggregate the output of multiple channels.The aggregation gate produces channel-wise weights dynamically, facilitating the fusion of resulting multiscale feature maps.This layout effectively enables the model to extract discriminative features even in challenging conditions.To evaluate the efficacy of our proposed MPRe-ID framework including body and Face features, we conducted experiments on the widely-used Market1501 and DukeMTMC-reID dataset.The experimental results compared with state-of-the-art approaches demonstrate the effectiveness of our MPRe-ID approach.