深度图优化之ThreadDepthCleaner

关注

原创最新推荐文章于 2026-11-07 21:30:32 发布 675 阅读 AI 写同款· GEO 优化 ›

【企业知识助手·Agent 实战】如何构建 Agent 运行态可观测体系：从行为基线漂移检测到成本预算治理的深度实战核心原理：运行态可观测的本质是把"单点阈值"升级为"与基线的分布距离"——埋点回答发生了什么，基线比较回答这是否还正常关键结论：deny 率没有单向好坏语义，必须双边告警 + known-good / known-bad 双哨兵仲裁；成本归因的终点是单位经济与预算三线一位；评测分数入库让质量拥有趋势线。阅读详情

github地址：ThreadedDepthCleaner

按github上的说明配置环境：

git clone https://github.com/juniorxsound/ThreadedDepthCleaner --recursive
mkdir build && cd build && cmake ../ && make -j4

这一步容易出现undefined reference to symbol ‘dlclose@@GLIBC_2.2.5‘
和undefined reference to symbol ‘pthread_create‘报错，
解决方法见：
undefined reference to symbol dlclose
undefined reference to symbol ‘pthread_create‘

github中的main.cpp是直接连接realsense相机的，这里没有用相机，直接优化深度图，
所以要改一下代码，最后cleanedDepth就是优化后的深度图。

Mat rawDepthMat = imread("depth.png", cv::IMREAD_UNCHANGED);
const int w = rawDepthMat.cols;
const int h = rawDepthMat.rows;

// Create an openCV matrix for the DepthCleaner instance to write the output to
Mat cleanedDepth(Size(w, h), CV_16U);

//Run the RGBD depth cleaner instance
depthc->operator()(rawDepthMat, cleanedDepth);

//补空洞
const unsigned char noDepth = 0; // change to 255, if values no depth uses max value
Mat temp, temp2;

// Downsize for performance, use a smaller version of depth image (defined in the SCALE_FACTOR macro)
Mat small_depthf;
resize(cleanedDepth, small_depthf, Size(), SCALE_FACTOR, SCALE_FACTOR);

// Inpaint only the masked "unknown" pixels
inpaint(small_depthf, (small_depthf == noDepth), temp, 5.0, INPAINT_TELEA);

// Upscale to original size and replace inpainted regions in original depth image
resize(temp, temp2, cleanedDepth.size());
temp2.copyTo(cleanedDepth, (cleanedDepth == noDepth));  // add to the original signal