EAPSI: Object Recognition for the Purpose of Traffic Compliance of Autonomous Vehicles

Campbell Joseph, Chandler AZ

Investigators

Abstract

In order for an autonomous vehicle to effectively navigate a road network, it needs to have accurate information about road elements such as traffic signs, traffic lights, and road markings so that it can comply with local traffic laws. Traditionally this information is gathered through detailed maps or visual recognition, however, these approaches can fail if there is insufficient map data available or on-board sensors are unsuccessful at locating road elements. This project proposes to build a system which can detect these road elements by analyzing the behavior of nearby road vehicles. This research will be performed under the guidance of Dr. Marcelo H. Ang Jr. at the National University of Singapore. Dr. Ang previously co-authored related research and as such will be a source of invaluable expertise, in addition to being a member of a lab with access to a cutting-edge autonomous vehicle platform. Nearby vehicles will be detected by using a combination of LIDAR sensors and cameras which are mounted to an autonomous vehicle platform. Vehicle positions will be extrapolated from the sensor data in conjunction with localization data provided by the vehicle platform, from which features will be extracted and fed into a machine learning classifier. Recent research has shown great success at identifying pedestrian paths by filtering out noise in pedestrian positions via clustering then modelling with a Naive Bayes Classifier, so these techniques will be used in order to identify road elements that lie on a vehicle?s path. The results from the classifier will be used in conjunction with other traditional identification methods in order to improve the overall identification rate of road elements. This NSF EAPSI award is funded in collaboration with the National Research Foundation of Singapore.