Pedestrian recognition and obstacle avoidance for autonomous vehicles using raspberry Pi

Abstract

The aim of this paper is twofold: firstly, to use ultrasonic sensors to detect obstacles and secondly to present a comparison of machine learning and deep learning algorithms for pedestrian recognition in an autonomous vehicle. A mobility scooter was modified to be fully autonomous using Raspberry Pi 3 as a controller. Pedestrians were initially simulated by card board boxes and further replaced by a pedestrian. A mobility scooter was disassembled and connected to Raspberry Pi 3 with ultrasonic sensors and a camera. Two computer vision algorithms of histogram of oriented gradients (HOG) descriptors and Haar-classifiers were trained and tested for pedestrian recognition and compared to deep learning using the single shot detection method. The ultrasonic sensors were tested for time delay for obstacle avoidance and were found to be reliable at ranges between 100 cm and 500 cm at small angles from the acoustic axis, and at delay periods over two seconds. HOG descriptor was found to be a superior algorithm for detecting pedestrians compared to Haar-classifier with an accuracy of around 83%, whereas, deep learning outperformed both with an accuracy of around 88%. The work presented here will enable further tests on the autonomous vehicle to collect meaningful data for management of vehicular cloud.