Pedestrian and Vehicle Detection for Visually Impaired People

Abstract

This paper proposes a method for helping visually impaired people by detecting and classifying incoming obstacles into pedestrians and vehicles on the road. When walking on roads or pathways, visually impaired persons have limited access to information about their surroundings; therefore, recognizing incoming pedestrians or cars is critical for their safety. Walking from one location to another is one of the most difficult tasks for visually impaired persons. White canes and trained dogs are the most often utilized instruments to assist visually impaired people in traveling and navigating. Despite their popularity, these technologies cannot offer the visually impaired people all of the information and functionality that persons with sight have access to and these aids are not that good for safe mobility. This proposed model aims to help visually impaired people by solving this problem. This paper proposes a model that can detect and classify pedestrians and vehicles on road using machine-learning and computer vision approaches. The system compares different classifiers based on SIFT and ORB feature extraction to evaluate the best approach. Different classifiers such as Random Forest, Decision Trees, SVM (3 kernels), and KNN are compared on the basis of testing accuracy, F1 score, recall, precision, sensitivity, and specificity. This study concluded that Random Forest yields the best result with 87.58% testing accuracy with SIFT feature extraction.