Road Detection System using Machine Learning Approaches

• Main Author: Winarto
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/42839
• Institution: Institut Teknologi Bandung

Detecting road area that of an image becomes an important research topic and is mostly done by researchers. Road detection is an essential requirement when implementing automated driving system. Road detection methods vary between heuristicbased approach and machine learning approach. Problems that are often encountered in detecting roads can be divided into road area problen (lighting, puddle, damaged road, shadows) and roadside problems (road markings, grass, soil, sidewalks). This final project investigates different machine learning techniques such as KNN, SVM, and Random Forest and different set of features to produce the best performing classifier. There are two types of classifier chosen and compared, namely binary classifier and 3-class classifier. Binary classifier is used to classify road and background and 3-class classifier is used to classify road, background, and roadside. The tested features to produce this classifier are color, neighbors color, normalized position, and linear binary pattern. The best-performing classifier is tested on several problems that exist in road detection. The classifier tested is Random Forest binary classifier with color and normalized position features. This classifier is tested to classify images and videos. From testing, the value of F1-Score is 0.944 and the average FPS is 20.5. The prototype of the developed road detection system has good performance in carrying out road classification. The conclusion is that the prototype system developed can classify road areas in an image or video.