MISSING CHILD IDENTIFICATION SYSTEM USING DEEP LEARNING AND MULTICLASS SVM

Abstract

Every year, a vast number of children in India are reported as missing. A significant proportion of missing kid situations result in the youngsters being untraced. This research introduces a new use of deep learning techniques to find missing children based on photographs of a large number of youngsters, using facial recognition technology. Members of the public have the ability to submit images of potentially suspect children to a shared online platform, along with accompanying details and descriptions. The picture will undergo an automated comparison with the recorded images of the missing kid stored in the repository. The supplied kid photograph is classified and the photo with the closest match will be picked from the database of missing children. In this process, a deep learning model is trained to accurately recognize the missing kid by using the face picture submitted by the public and comparing it with the missing child image database. Face recognition in this context utilizes the Convolutional Neural Network (CNN), a powerful deep learning method specifically designed for image-based tasks. Face descriptors are obtained from the photos by using a pre-trained Convolutional Neural Network (CNN) model called VGG-Face, which is based on a deep architecture. Our technique differs from typical deep learning applications in that it use a convolutional network just as a means of extracting high-level features. The actual detection of children is then performed by a trained SVM classifier. By selecting the most effective CNN model, VGG-Face, and properly training it, a deep learning model can be created that is unaffected by noise, illumination, contrast, occlusion, image pose, and the age of the child. This model surpasses previous methods in identifying missing children based on face recognition. The kid identification system had a classification performance of 99.41%. The evaluation was conducted on 43 instances involving children.