Machine Learning Approaches on Pedestrian Detection in an autonomous vehicle

Abstract

In autonomous driving, detecting pedestrians is a safety-critical activity, and the decision to avoid a person must be made as quickly as possible with as little delay as possible. In this work, INRIA and PETA datasets are taken. The progression of the work that is being proposed is broken up into three phases. The first step is to detect edges, the second step is to group colours, and the third step is extracting the feature, which includes screening body parts of pedestrians and detecting shoulder lines. The machine learning classifiers such as SVM, Naïve Bayes and KNN are taken for predicting the pedestrian in the road. The accuracy for SVM, Naïve Bayes and KNN are calculated as 93.58, 94.42 and 98.44 respectively. With the KNN model, it achieves the highest accuracy for predicting the exact images.