A DESIGN OF AUTONOMOUS REMOTE CONTROL CAR USING CONVOLUTIONAL NEURAL NETWORK AND HAAR-LIKE FEATURES CLASSIFIER

Abstract

Autonomous vehicle is a type of vehicle that can drive safely without any human intervention. This safety is related with the capability of the vehicle to keep driving on its own track without disturbing the other lines, detect objects in front of it and estimate the distance to that object to prevent accident. However, there are only few researchers developing autonomous vehicle that can follow the predetermined path, detect objects, and estimate distance to said objects. In this final project, the author wants to make an autonomous remote control car that has those three features. This project develops autonomous remote control car that is controlled by convolutional neural network to keep the car in track. The device has features to classify three object classes (i.e. pedestrian, car, and stop sign) by using Haar-like classifier. In addition, the device can estimate the distance to the object by using pinhole imaging theory. The device takes images from a mobile phone attached to the car as its only input and processes the images in MATLAB2019a. This final project describes the entire process of its creation from hardware requirements, through the design of the control system, up to the selection and training of a convolutional neural network. The final device is able to follow the track with the accuracy ranging from 86.67% to 100.00% and classify three object classes with the accuracy ranging from 53.33% to 86.67%. In addition, the device can estimate the object distance with average error equals to 2.43 cm.