基于多传感器融合的复杂越野环境人员识别方法

军用无人车辆的一大应用场景是班组伴随功能,复杂越野环境下的人员识别技术,是军用无人车辆的关键技术,实时、准确地检测出车辆周边的人员,是班组伴随实现的前提条件。针对激光雷达和摄像头融合感知下的人员识别问题,提出基于多传感器融合的解决方案,针对激光雷达采用基于KDTree加速的欧式聚类方法,针对摄像头信息设计改进的YOLO v3深度学习网络架构,设计空间尺度融合算法。通过履带式车辆验证平台在复杂环境下进行多工况试验验证,结果表明,所设计的基于多传感器融合的识别算法能够准确识别复杂环境下的目标,激光雷达和摄像头融合算法的交并比超过95%。

Person Identification in Complex Off-road Environments Based on Multi-Sensor Fusion

Accompanying the team was a major application of military unmanned vehicles. Thereforereal- time person identification with high accuracy in the complex off-road environment was the key technology for military unmanned vehicles. In this papera method of person identification based on LiDAR and camera fusion was proposed. The KDTree-accelerated European clustering method was used for the LiDAR system and the improved YOLO v3 deep learning network architecture was designed to process the live stream from the camera, and then the fusion criteria for different spatial scales were determined.The experiments were conducted on a verification platform for crawler vehicles under different working conditions.The results show that the proposed identification algorithm based on multi- sensor fusion can accurately identify the target in complex environment, and the intersection over union exceeds 95% for the LiDAR and camera fusion algorithm.