非结构化环境下无人驾驶车辆跟驰方法

A kind of multi-sensors fusion method was proposed for autonomous vehicle following application in unstructed environment to solve the problem of the guide vehicle locating in autonomous vehicle following task, and to obtain relative position and speed of guide vehicle in real time. Firstly, oriented by the task requirements, the sensors selection work was completed in term of unstructed environment features by on board experiments results. Then, a joint calibration work-flow was designed for the selected millimeter wave radar (MMW) and camera to realize data spatial fusion. And several works were carried out, including designing the targets distance related transversal distance constraint threshold, analyzing the target data reliability based on annals frame to solve radar data jump and virtual detection phenomenon in unstructed environment and to accomplish radar valid targets extraction, creating the candidate vehicle detection boxes with changeable area based on camera pin-hole imaging principle. Finally, an output information procedure was designed with regard to vehicle following application, based on the mainstream deep learning framework about object detection. On board test results indicate that, the proposed method can satisfy the basic requirements of vehicle following application in most off-road scenes with a certain output stability and accuracy.