Novel Real-time Safety Algorithm for Predicting Danger-ous Targets on the Driving region in the Unstructured Road

Abstract

A certain number of people die every year from traffic accidents on unstructured farmland roads. Detection and prediction of dangerous targets could help to reduce accidents. A novel real-time algorithm based on YOLACT, a combination of GPM and RCM, was proposed to achieve the goal. On this basis, a self-collected data set of 5000 test samples is used for testing. In addition, we have carried out experiments on actual roads. The results showed that, targets on the road could be detected, tracked and predicted. The detect accuracy of algorithm in the data set could reach up to 90%, the processing speed to 30.4 fps. This study can be used for automatic driving and assisted driving.