ELECTRIC WHEELCHAIR MOVEMENT CONTROL BASED ON FINGER PATTERN RECOGNITION WITH CONVOLUTIONAL-LSTM METHOD

Abstract

Wheelchairs are a tool for people with disabilities who have difficulty walking to do their daily activities. Users of conventional wheelchairs will quickly get tired if they have to walk long distances. The application of technology has expanded in all fields including biomedical. With the advancement of technology, a variety of wheelchair control interfaces have been developed.In this study, the researchers will develop on the navigation control of the movement of the electric wheelchair based on the identification of the finger posture of the hand. Five finger patterns represent the motion of the wheelchairs to move forward, backwards, right, left, and stop. The research data set was successfully collected as much as 8000 data per finger pattern with a total of 40,000 samples. The proposed methods by the researchers are using deep learning methods such as Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM). The combination of CNN and LSTM was chosen because the CNN method is highly focused on solving current object recognition problems. The combination with the LSTM method is expected to increase the degree of accuracy of the identification of five finger patterns that represent the movement of the wheelchair: forward, backward, right, left, and stop. Based on the results of offline testing, the accuracy score was 96.9% and the average response time was 179 ms. The highest accurate value in the advanced class was 99.7% and the lowest accurately in the stop and right classes was 97.9%. The greatest recall or sensitivity value at the stop class is 99.6% and the least recall value on the left class is 9.7%. The system is capable of predicting as much as ±6 frames per second. (fps).

Abstrak

Kursi roda merupakan alat bantu bagi penyandang disabilitas yang mengalami kesulitan berjalan untuk melakukan aktifitas sehari-hari. Pengguna kursi roda konvensional akan cepat lelah jika harus berjalan jauh. Penerapan teknologi telah merambah pada semua bidang termasuk dalam bidang biomedis. Seiring dengan perkembangan teknologi, telah dikembangkan berbagai antarmuka kontrol kursi roda.Pada penelitian ini peneliti akan melakukan pengembangan pada kontrol navigasi pergerakan kursi roda elektrik berdasarkan pengenalan pose jari tangan. Lima pola jari tangan mewakili pergerakan kursi roda untuk bergerak maju, mundur, kanan, kiri, dan berhenti. Dataset penelitian yang berhasil dikumpulkan sebanyak 8000 gambar data tiap pola jari tangan dengan total sampel sebanyak 40000 gambar. Metode yang peneliti usulkan menggunakan metode deep learning yaitu Convolutional Neural Network (CNN) dan Long Short-Term Memory (LSTM). Metode gabungan CNN dan LSTM  dipilih karena metode CNN sangat mumpuni untuk menyelesaikan permasalahan pengenalan objek saat ini. Penggabungan dengan metode LSTM diharapkan menambah tingkat akurasi dari pengenalan lima pola jari tangan yang mewakili pergerakan kursi roda yaitu maju, mundur, kanan, kiri, dan berhenti. Berdasarkan hasil pengujian offline didapatkan nilai akurasi sebesar 96,9% dan rata-rata time respon sebesar 179 ms. Nilai presisi paling besar pada kelas maju sebesar 99,7% dan nilai presisi paling kecil pada kelas berhenti dan kanan sebesar 97,9%. Nilai recall atau sensivitas paling besar pada kelas berhenti sebesar 99,6% dan nilai recall paling kecil pada kelas kiri sebesar 96,7%. Sistem yang dibuat mampu melakukan prediksi sebanyak ±6 frame per second (fps).