An improved traffic lights recognition algorithm for autonomous driving in complex scenarios

Image recognition is susceptible to interference from the external environment. It challenging accurately and reliably recognize traffic lights in all-time all-weather conditions. This article proposed an improved vision-based algorithm for autonomous driving, integrating deep learning multi-sensor data fusion assist (MSDA). We introduce a method obtain best size of region interest (ROI) dynamically, including four aspects. First, based on (RTK BDS/GPS, IMU, camera, LiDAR) acquired normal environment, we generated prior map that contained sufficient information. And then, by analyzing relationship between error sensors optimal ROI, adaptively dynamic adjustment (ADA) model was built. Furthermore, according positioning ADA model, ROI can be obtained predict location lights. Finally, YOLOv4 employed extract identify image features. evaluated our using public set actual city road test at night. The experimental results demonstrate has relatively high accuracy rate complex scenarios promote engineering application driving technology.