This procedure implemented by C plus plus with Open Soure Library OpenCV, PCL, and Realsense Developer SDK.

Intel® RealSense™ camera fits remarkable technology into a small package. There are three cameras that act like one—a 1080p HD camera, an infrared camera, and an infrared laser projector—they “see” like the human eye to sense depth and track human motion. Intel® RealSense™ technology1 redefines how we interact with our devices for a more natural, intuitive and immersive experience, supported by the powerful performance of Intel® processors.

And you can find more details with Realsense on https://software.intel.com/en-us/intel-realsense-sdk/download

This program consists of four parts: Segmentation, classification, registration and reflection. First, depth data can be divided into serval regions with improved region-growing segmentation. Then classifly region catergory with a classic linear svm with HOG as feature. Next, regist mesh generated by Realsense with 3D-model which we built by 3D scanner inadvance. Last but not least, reflect Points Cloud to 2D-color-image. Finally, we can see purple point in folowing picture.

An improved region-growing segmentation efficiently divide depth data into topk main regions.

An classic Support Vector Machine (SVM) with Histogram of Oriented Gradient (HOG) as feature rapidly select regions contained objects.

Regist Point Cloud (3D mesh) with pre-built 3D-model, by RANSAC and ICP.

Reflect points into 2D image.

color and depth are soure data captured by Realsense. before merging is raw result with region-growing and segmentation merge small regions into representive regions as main regions. regions show the bounding boxes with convex hulls of main regions. And classification show region which probably contains object. The 3D Point Cloud window show the registration and reflect demonstrate the final result.

We just add some solutions which we have finished off into Issues. Welcome You to push problems you faced with.