A Visual C++ project to test run the OpenCV DNN object detection with RealSense camera. The GUI utilizes a modified version of gorgeous nanogui and use OpenGL shader to render video streams.

The Deep Neural Network part of codes are derived from MobileNet Single-Shot Detector example of OpenCV. The trained Caffe model (included in project resources) is loaded at the beginning and used to classify 20 object classes from RGB image. Note that only the center square ROI is the hot detecting spot, the unused side bands shall turn gray by design. For each detected object, the distance is estimated from its detected location within corresponding depth frame.

A test run video captured from screen is available on YouTube here.

This VC++ project rely on the integration environment of vcpkg ports. Follow the instruction to install vcpkg and run .\vcpkg integrate install, then install following dependency ports with .\vcpkg install <port_name>:x64-windows:

• opencv
• realsense2
• poco
• eigen3
• stb
• glfw3

, where eigen3, glfw3, and stb ports need to setup additional static liked packages for nanogui to build. Please create a file "x64-windows-static-md.cmake" in vcpkg/triplets folder with the following contents,

set(VCPKG_TARGET_ARCHITECTURE x64)
set(VCPKG_CRT_LINKAGE dynamic)
set(VCPKG_LIBRARY_LINKAGE static)

, then install these three ports with .\vcpkg install <port_name>:x64-windows-static-md.