Ego Lane Yaw Rate Extraction Using LaneNet Network

Evolution of transportation has rapidly grown with the existence of autonomous navigation technology in the world. Good navigation comes with a high accuracy localization system. The current localization techniques are always prone to solidity and availability of extracted features. These occurrences might have limitations to places where it has limited features and wide area to be extracted such as on highways. This situation could bring error to the localization system and the autonomous vehicle (AV) may cause fatality to the people around it. Hence, an alternative localization method needs to be implemented which makes use of the non-changing features and available in all roads in the world which is the road lane marking information. To integrate the AV localization system with the road lane information, the vehicle first needs to extract the yaw orientation of the detected lane to predict the pose and orientation estimation based on the curvature of the road and slope of the road lane observed. Therefore, this paper proposed a yaw rate extraction method onto LaneNet network to extracts the road lane using a High Definition (HD) camera. This experiment is conduct in two different frames which are in the local frame (vehicle coordinate frame) and in the global frame (UTM coordinate frame) and the result are compared. The yaw rate extracted in local frame is the best solution if compared to yaw rate extraction in global frame due to the transformation coordinates into global coordinates are exposed to tolerance error which possibly cause by multi-path error or noise interruption at atmospheric layer.