AuTinho

0 0
  • 0 Collaborators

AuTinho is a module for the recognition of emotions using machine learning techniques to assist in the analysis and study of the emotions of children with ASD, to create an activity plan focused on the preferences of the child and increase the effectiveness of the treatment. ...learn more

Project status: Under Development

Robotics, Artificial Intelligence

Intel Technologies
DevCloud, Intel FPGA

Links [2]

Overview / Usage

Autistic Spectrum Disorder (ASD) is a disorder of the neurodevelopment that affects communication, social interaction, how to express emotions, among others. Understanding the emotions expressed during a therapeutic session helps to create an activity plan focused on the child's preferences in order to increase the effectiveness of the treatment. This work proposes the development of a module for the recognition of primary emotions using machine learning techniques to assist in the analysis and study of the emotions of ASD patients. The module will be based on Convolutional Neural Network Models that will be implemented, analyzed and executed on the Openvino Starter Kit platform, which has an FPGA Cyclone 5.

Methodology / Approach

The first step is the creation of a model to perform the tasks of face detection and emotion recognition, it is intended to use a model based on Convolutional Neural Networks, which offers good precision in many applications related to machine learning. The training and validation of the model will be done using the AffectNet dataset, which is the largest database of facial expressions, which allows searches in the automated recognition of facial expressions. The images in this bank are classified into 8 different types of emotions. But for the purpose of this work, only the six emotions described above, which represent the basic emotions, will be needed.

The second step will be the implementation of the model on the Openvino Starter Kit platform, which includes an FPGA Cyclone 5. For programming the platform, an Upsquared card will be used, necessary to store the validation images, read the dataset images through functions of the OpenCV and store these images in the FPGA memory.

Finally, the metrics used for evaluation will be the average frame processing time (fps) and the accuracy in recognizing emotions. After analyzing these results, it will be possible to study other models that may be better suited to the problem.

Technologies Used

  • TensorFlow
  • Intel DevCloud
  • FPGA Cyclone 5
  • Upsquared
  • Uses OpenVINO as a ML model optimizer

Comments (0)