Comparison of SVM, Random Forest and KNN Classification By Using HOG on Traffic Sign Detection

Abstract

A lot of traffic accidents caused by traffic violations gave rise to an idea to develop an Advanced Driver Assistance System (ADAS) technology. Challenge in developing ADAS is how the system can recognize the environment on the highway, one of which is Traffic Sign Detection and Recognition (TSDR). Therefore, this study proposes a traffic sign detection model on German Traffic Sign Detection Benchmarks (GTSDB) dataset using HSV color segmentation and combined with Maximally Stable Extremal Regions (MSERs). In addition, morphological processes are carried out to reduce noise, with erosion and dilation followed by the calculation of the area and calculation of the aspect ratio. Color segmentation and morphological processes produce Region of Interests (ROIs). For each ROIs, feature extraction is performed using the Histogram of Oriented Gradients (HOG). To determine whether an ROIs is a traffic sign or not, classification is carried out using Support Vector Machine (SVM), Random Forest and K Nearest Neighbors (KNN). The accuracy results are HOG+SVM 79.05%, HOG+Random Forest 79.64%, and HOG+ KNN 81.65%.