Claim Missing Document
Check
Articles

Found 1 Documents
Search
Journal : Journal Of Artificial Intelligence And Software Engineering

The Detection of Objects and Distance for the Visually Impaired by Using Deep Learning ResNet-152 and the Triangulation Method Ichwan, Muhammad; Dewi, Irma Amelia; Salsabilla, Nadiati
Journal of Artificial Intelligence and Software Engineering Vol 5, No 1 (2025): Maret
Publisher : Politeknik Negeri Lhokseumawe

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30811/jaise.v5i1.6419

Abstract

This research aims to detect objects and determine the distance from the mobile camera to facilitate and assist visually impaired users in recognizing the surrounding environment using several models made with the RetinaNet Method and Residual Network-152 architecture. Three object detection models were generated by ADAM and SGD parameter optimizers. Object recognition was performed using the TensorFlow framework with a dataset of 2,444 images. The first model with training parameters used is ADAM optimizer, epoch 50, batch size 16, and lr 1e-5. The second model has training parameters, such as ADAM optimizer, epoch 100, batch size 16, and lr 1e-5. The third model uses SGD optimizer training parameters, epoch 50, batch size 16, and lr 1e-5. Based on 250 tests on each model, the results show that the best model is the first model, which shows a precision value of 82%, a recall value of 98%, an f1 score value of 89%, and an accuracy value of 86%. The distance from the mobile camera is tested in multiples of 10 at a distance of 100-300 cm with a camera height of 100-130 cm and a camera angle of 80⁰-90⁰ getting reasonable distance detection results at a camera height of 130 cm because it gets the smallest total difference value of 14.3 cm.