Build a DIY security camera with neural compute stick - Tutorial

Also check out the more advanced version achieves 8.69 fps real-time object detection on Raspberry Pi using FIFO, video_objects_threaded.

Introduction

This project uses SSD MobileNet to do object recognition and classification for a webcam. The companion Arduino sketch can be downloaded from repo CamGimbal.

The provided Makefile does the following:

1. Builds both caffe ssd mobilenet graph file from the caffe/SSD_MobileNet directory in the repository.
2. Copies the built NCS graph file from the SSD_MobileNet directory to the project base directory
3. Downloads some sample traffic video files.
4. Runs the provided street_cam_ssd_mobilenet.py program which creates a GUI window that shows the video stream along with labels and boxes around the identified objects.

Prerequisites

This program requires:

• 1 NCS device
• NCSDK 1.11 or greater
• opencv 3.3 with video for linux support

Note: The OpenCV version that installs with the current ncsdk (1.10.00) does not provide V4L support. Check the Part 1 tutorial to "Install OpenCV3 the easy way"

Note: All development and testing has been done on Ubuntu 16.04 on an x86-64 machine as well as Raspbian Stretch on Raspberry Pi 3 Model B.

Makefile

Provided Makefile has various targets that help with the above mentioned tasks.

make help

Shows available targets.

make run_cam

Runs the provided python program which shows the webcam live video stream along with the object boxes and classifications. Save 'person' images to folder images.

make run_gimbal

Runs the provided python program which shows the webcam live video stream along with the object boxes and classifications. Save 'person' images to folder images. Connect to Arduino serial port turning a servo motor to follow one or more detected persons. Check part 2 of the tutorial on how to build one yourself.

make all

Builds and/or gathers all the required files needed to run the application except building and installing opencv (this must be done as a separate step with 'make opencv'.)

make videos

Downloads example video files.

make run_py

Runs the provided python program which shows the video stream along with the object boxes and classifications.

make clean

Removes all the temporary files that are created by the Makefile