Construction of Driving image dataset under Adverse conditions for Traffic light recognition in Urban Automated Driving

Abstract

In automated driving in urban areas, the vehicle must be able to operate the safety driving following the traffic rules based on the onboard sensors data and the predefined digital map. When the vehicle approaches an intersection, it is necessary to determine safety based on the status of surrounding traffic participants and traffic lights recognized by the onboard sensors. Especially, the lighting color of traffic lights must be recognized using only the onboard camera. However, there are situations in which it is difficult to recognize the traffic lights based on images alone due to changes in sunlight and weather conditions. In our research project, we have been conducting demonstration experiments of automated driving on Japanese public roads and discussing the limitations of autonomous recognition using onboard sensors. Then, we have developed an image dataset of intersection driving including challenging image conditions for traffic light recognition obtained through our demonstration experiments.