y27choi's profile - activity

I'm using Intel RealSense R200 camera to capture 3D information to update the costmap for move_base on Ubuntu 14.04 LTS.

Unfortunately, the pointcloud data produced from the depth stream is too noisy to be used to plan a proper path. Does anyone know an effective, less computationally expensive, noise removal algorithm for the depth stream?

I have tried OpenCV's median filter and bilateral filter, as well as the PCL noise removal filter which seem to help a little but didn't remove the noise entirely - plus they were all computationally expensive with a bigger kernel size setting. Also, most of the filters required 8 bit depth or floating points as well, but the depth stream from RealSense ROS has a 16-bit depth format. Hence, extra processing to convert 16-bit depth to floating point was required.

I have seen many people using this camera without any problem and it will be really appreciated if someone can help me with this noise issue!