基于深度学习理论,对电子海图与雷达图像船舶感知信息融合进行了研究,通过三维点云数据特征将满足要求的点云筛选出,并将点云数据在二维平面投影进行聚类。支持向量机(SVM)分类器和视觉显著性两种方法结合跟踪船只,采集正负样本并进行正负样本的方向梯度直方图特征提取,完成SVM分类器训练;通过训练好的SVM分类器识别视频图像的目标船只,利用亮度、颜色等一致性特点检测此感兴趣区域的显著性。分析了视觉显著性、激光雷达、SVM分类器检测实施船只的追踪和位置定位功能,将视觉显著性、激光雷达、SVM分类器三者之间存在的相应转化关系进行分析研究,从而得到激光雷达系统和拍摄相机间的联动性能,并予以连接,最终完成电子海图和激光雷达影像,传感器在进行数据上传的同时,可以实时进行。实验表明:电子海图和雷达图像融合检测真正率为97.14%。

Based on the deep learning theory, fusion of ship perception information of electronic chart and radar image is studied, the point clouds that meet the requirements are selected through three-dimensional point cloud data features, and the point cloud data is cluttered in two-dimensional plane projection. Support vector machine (SVM) classifier and visual saliency are combined to track ships, collect positive and negative samples and extract their histogram of oriented gradient features to complete the training of SVM classifier; the trained SVM classifier is used to identify the target ship of video image, and the saliency of interested region is detected by the consistency characteristics of brightness and color. Functions of visual saliency, lidar and SVM classifier in ship tracking and positioning, and the corresponding transformation relationship among visual saliency, lidar and SVM classifier are analyzed, so as to obtain the linkage performance between lidar system and camera, connect them, and finally achieve the electronic chart and lidar imaging technology. In the process of data upload and transmission, the sensor can achieve real-time operation. Experiments show that the real rate of fusion detection of electronic chart and radar image is 97.14%.

TN957.51

山西省教育科学规划基金项目