关键词: 隧道环境, 无人驾驶, 多传感器融合, D-S证据理论, “目标决策”两级融合策略

Abstract: Based on the special driving environment in the tunnel and the perception requirements of unmanned driving, appropriate sensors and hardware were chosen to build a test vehicle and a perception system of multi-sensor fusion of millimeter-wave radar and camera. A two-level information fusion algorithm of “target-decision” was proposed based on YOLOv4 target-level information fusion algorithm and improved D-S evidence theory. Finally, a verification test of perception information two-level fusion was carried out in the urban road tunnel environments. The results show that in the tunnel environments, compared with the single camera or the millimeter-wave radar sensing results, the target-level fusion result based on the association of the camera and the millimeter-wave radar sensor to perceive the ROI area may improve the recognition accuracy by 9.51%, making up for the shortcomings of a single sensor in the tunnel environment perception technology. Based on the target-level fusion perception results, using the improved D-S evidence theory algorithm to perform decision-level fusion, compared with the single target-level fusion results, the false detection rate is reduced by 3.61%, which significantly improves detection accuracy. By adopting the multi-sensor sensing information target-decision-making two-level fusion strategy, it may meet the reliable sensing requirements of unmanned vehicles in the special tunnel environments, and provide theoretical and technical support for promoting the applications of unmanned controlled technology.

Key words: tunnel environment, unmanned driving technology, multi-sensor fusion, D-S evidence theory, “target-decision” two-level fusion strategy