Face Search AI: Wonderla Park Visitors Photo Search and Monitoring Using Hybrid CNN & LSTM Approaches

Amudala Jaswanth1, Dr. N.Deepak Kumar2, Dr.M.Giri3

1P.G Scholar, Department of CSE, Sree Rama Engineering College, Tirupati, Andhra Pradesh, India, jaswanthamudala82@gmail.com

2Professor, Department of Computer Science & Engineering, Sree Rama Engineering College, Tirupati, Andhra Pradesh, India, hod.cse@sreerama.ac.in

3Professor, Department of CSE (AI), Mother Theresa Institute of Engineering and Technology, Palamaner, Andhra Pradesh, India, dr.m.giri.cse@gmail.com

Abstract -Millions of people are interested to visit Wonderla Park, the Park is located in four different branches, visitors are allowed to visit at different places with different time internals, visitor’s photos are captured by the Park management, and that will create large volume of photo suite. Traditional Approaches mostly relay on timestamp of picture captured, ids are allocated based on tickets, tagging is applied to pictures manually, and fail to identify a group of sequence images of a particular visitor.In this paper we designed automatic FS-AI (Face Search AI), an innovative Wonderla Park images discovery, monitoring, and management approach by combining both CNN (Convolutional Neural Network) and LSTM (Long Short Term Memory) technique, CNN based technique is capturing the visitors images by analyzing their face expressions (happy mood), all temporal are modelled with help of LSTM approach to capture continuous images when visitor face feeling changed, same visitor images are gathered from multiple locations and then group then under single visitor ID. Data samples are collected from various branches of Wonderla Park that are located all over India in different states.Experiments are conducted with help of different metrics and with different face discovery techniques. Performance values of FS-AI face discovery and grouping approach is evaluated, results of FS-AI method compared with similar techniques like G-CNN, ResNet-50, VGG-16, H-SVM, LBPH, S-CNN, and EF approaches. FS-AI method exhibits lower EER value, shows lower ART value, achieved greater accuracy value, exhibits greater precision value, exhibits greater recall value, shows greater F1-score value, achieved greater mAP value, exhibits greater AUC value, and from the results proposed FS-AI face discovery and grouping approach shows greater performance than other similar kinds of techniques.

Key Words: Machine Learning, Deep Learning, CNN,G-CNN, ResNet-50, VGG-16, H-SVM, LBPH, S-CNN, image classification, Grouping of images, and EF