使用PCL合并点云时触发Segmentation fault错误的修复咨询

场景描述 #

从图像序列提取兴趣点，使用PCL库拼接所有关键帧点云生成全局地图。p是单关键帧提取并变换后的点云，globalMap是存储全局点云的智能指针，执行*globalMap += *p时触发段错误。

相关代码 #

generatePointCloud(pKF);
pcl::PointCloud<pcl::PointXYZRGBA>::Ptr p(new pcl::PointCloud<pcl::PointXYZRGBA>);
// Eigen::Matrix<double,4,4> Converter::toMatrix4d(const Sophus::SE3f &T)
pcl::transformPointCloud(*(pKF->mptrPointCloud), *(p), Converter::toMatrix4d(pKF->GetPoseInverse()));
{
    std::unique_lock<std::mutex> lck(mMutexGlobalMap);
    std::cout<<"MutexGlobalMap "<<std::endl;
    if (globalMap == nullptr) {
        std::cout<<"NUll pointer "<<std::endl;
    } else {
        std::cout<<"not empty "<<std::endl;
    }
    if (p == nullptr) {
        std::cout<<"NULL POINTER  "<<std::endl;
    } else {
        std::cout<<"not empty "<<std::endl;
    }
    if (p->size() > 0) {
        std::cout<<" P size "<<std::endl;
        cout<< p->size() <<endl;
    } else {
        cout<< p->size() <<endl;
    }
    if (globalMap->size() > 0) {
        std::cout<<" global map size "<<std::endl;
        cout<< globalMap->size() <<endl;

    } else {
        cout<< globalMap->size() <<endl;
    }

    // Check the type of globalMap
    std::cout << "Type of globalMap: " << typeid(*globalMap).name() << std::endl;
    // Check the type of p
    std::cout << "Type of p: " << typeid(*p).name() << std::endl;
    *globalMap += *p;
}

调试输出 #

Start processing sequence ...
Images in the sequence: 5178

First KF:0; Map init KF:0
New Map created with 575 points
receive a keyframe, 0
receive a keyframe, 1
receive a keyframe 
MutexGlobalMap 
not empty 
not empty 
 P size 
13415
0
Type of globalMap: N3pcl10PointCloudINS_12PointXYZRGBAEEE
Type of p: N3pcl10PointCloudINS_12PointXYZRGBAEEE
Added the point to the global map
tmp init
Setting Input Cloud
13415
receive a keyframe, 2
not empty 
not empty 
 P size 
13346
 global map size 
13415
Type of globalMap: N3pcl10PointCloudINS_12PointXYZRGBAEEE
Type of p: N3pcl10PointCloudINS_12PointXYZRGBAEEE
Segmentation fault (core dumped)

错误栈信息 #

Thread 4 "rgbd_tum" received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7fffd75b7700 (LWP 72283)]
__GI___libc_free (mem=0xb6934) at malloc.c:3102
3102    malloc.c: No such file or directory.

修复方案 #

1. 确保globalMap的有效初始化
   确认globalMap是通过合法方式创建的智能指针，例如：

   globalMap = pcl::PointCloud<pcl::PointXYZRGBA>::Ptr(new pcl::PointCloud<pcl::PointXYZRGBA>);
   

   避免使用未初始化的野指针，或指向已释放内存的失效指针。

2. 严格保证线程安全
   所有对globalMap的操作（包括初始化、重置、修改、读取）都必须在mMutexGlobalMap锁的保护下执行，防止多线程竞态导致指针状态异常。

3. 替换点云拼接方式
   改用insert方法替代+=运算符，避免PCL运算符内部潜在的内存管理问题：

   globalMap->insert(globalMap->end(), p->begin(), p->end());
   

4. 验证pKF->mptrPointCloud的有效性
   检查generatePointCloud(pKF)是否正确初始化了pKF->mptrPointCloud，确保该指针指向合法的点云内存，避免transformPointCloud写入非法内存破坏堆结构。

5. 启用PCL调试模式排查
   编译时添加-DPCL_DEBUG宏，输出PCL内部内存管理的详细日志，定位具体的内存错误点。

内容的提问来源于stack exchange，提问作者Iscelia