Accurate Age and Gender Prediction Using DNN Model from Real World Camera Feeds

Abstract

Introduction: The enormous rise of video and image data’s has created a tremendous need for intelligent systems to autonomously understand and analyse information, while human interpretation is difficult. In social interactions, a person's face is crucial for expressing their identity and emotions.  Compared to machines, humans are not very good at distinguishing between distinct faces.Predicting age and gender from live video broadcasts is a difficult problem with many potential uses. Traditional approaches are unable to detect faces with any degree of accuracy due to the wide variation in facial images.

Objectives: To develop a humanoid systems employing Deep Neural Network (DNN) with improved preprocessing and overfitting prevention for predicting the age and gender accurately..

Methods: The study uses a TensorFlow DNN for estimating the the age and gender of a person. The data processing including cleaning and data augmentation ensures the model’s optimal performance. A Dropout Layer (DL) has been added to the feature extraction to prevent overfitting. The developed DNN model is evaluated using the face images in the Kaggle repository. Deep CNN (DCNN), Support Vector Machine (SVM), and K-Means clustering are compared with the suggested approach using the performance measures to determine the efficiency of the model.

Results: The findings show that, the TensorFlow DNN model outperforms the contrasted method with  an accuracy, precision, recall, and F1-Scores of about 98%, 96%, 95%, and 75%, respectively, showing its efficiency in age and gender predictions. Moreover, the face detection capability of the system allows for the development of a central suggestion system for humanoid robots, thus making them more functional and coordinated according to categorized age groups.

Conclusions: The research presents a TensorFlow DNN model optimized for real-time age and gender prediction from camera feeds.  The suggested model outperforms the algorithms such as DCNN, SVM, and K-Means clustering by embracing tedious data preprocessing methods, such as cleaning and augmentation, and leveraging benchmark datasets like IMDB and Adience.