基于多相机融合和特征点加速匹配的煤矿井下视觉同步定位与建图算法

A Visual Synchronous Location and Mapping Algorithm for Coal Mine Underground Based on Multi-camera Fusion and Feature Point Acceleration Matching

基于特征点法的视觉同步定位与建图(SLAM)算法在煤矿复杂环境下具有较多的应用场景,但随着图像质量以及相机数量的增加,特征点的匹配耗时会延长,相机与系统参数的适配复杂度也会提高。针对以上问题,提出一种基于多索引哈希和负位姿熵的视觉SLAM算法。该算法通过切分特征点描述子,查询存储图像信息的哈希表来提升匹配速度,不需要加载离线词典;同时根据信息熵理论,量化关键帧与当前地图的不确定度,确认关键帧的插入时机,简化传感器调参流程。试验结果表明,该算法相较于ORB-SLAM2算法,对于多特征点的视觉里程计,匹配速度获得了明显提高,且减少了内存占用空间;在复杂场景的数据集中,系统定位的精准度得到了提升,且降低了多相机系统的调参难度。

The visual synchronous localization and mapping(SLAM)algorithm based on feature point method has many application scenarios in complex coal mine environments.However,with the increase of image quality and the number of cameras,the matching time of feature points will be extended,and the complexity of camera system parameter adaptation will also increase.In response to the above issues,a visual SLAM algorithm was proposed based on multi-index hashing and negative pose entropy.This algorithm improves matching speed by segmenting feature point descriptors and querying the hash table that stores image information,without the need to load an offline dictionary.At the same time,based on information entropy theory,quantify the uncertainty between keyframes and the current map,confirm the insertion time of keyframes,and simplify the sensor parameter adjustment process.The experimental results show that compared to the ORB-SLAM2 algorithm,this algorithm significantly improved the matching speed and reduced the memory footprint for visual odometers with multiple feature points.In complex scene datasets,the accuracy of system positioning has been improved and the difficulty of parameter adjustment for multi camera systems has been reduced.