RANCANG BANGUN OBJECT DETECTION PADA ROBOT SOCCER MENGGUNAKAN METODE SINGLE SHOT MULTIBOX DETECTOR (SSD MOBILENETV2)

Abstract

Artificial intelligence or AI is a technology that emphasizes machine intelligence in responding
like humans, developed to help support human work. AI has widely applied in various fields
such as industry, medical, education, business, and robotics. The development of AI in the field
of robotics produces autonomous robots, one example is the KRSBI-Beroda soccer robot. This
research discusses the application of AI in the design of object detection systems using the
single-shot multibox detector (SSD) model on the KRSBI-Beroda soccer robot. This study aims
to produce an AI model in the form of an artificial neural network (ANN) implanted in the soccer
robot to distinguish between ball, goal, robot, and obstacle objects. This object detection system
was built using a deep learning method assisted by the TensorFlow Object Detection API
framework using the MobileNetV2 SSD model which is run using the python programming
language on the NVIDIA Jetson Nano board which is integrated into the C922 Pro webcam
camera. The model was built using a dataset of 977 images consisting of 3064 objects that
were trained as many as 200,000 steps on Google Colaboratory. The results showed a model
with an average mAP of 0.80 with an average total loss of 1.5. Validation of the model resulted
in a success rate of object prediction with an average accuracy of up to 98.45%.