[English] MobileNet-SSD

Ultra-fast MobileNet-SSD(MobileNetSSD) + Neural Compute Stick(NCS) than YoloV2 + Explosion speed by RaspberryPi.
Multiple moving object detection with high accuracy.
Video playback and object detection are executed asynchronously.
In order to realize high speed rendering with multi stick, it is implemented in multithreading/OpenGL.
【Warning】 This repository does not support NCS2.

【Japanese Article】
https://qiita.com/PINTO/items/b97b3334ed452cb555e2

【USB Camera + MultiProcessing High performance version】
Below, when using multiple sticks, it is more than three times the performance of this repository program.
I recommend that you refer to the following repository.
https://github.com/PINTO0309/MobileNet-SSD-RealSense.git

Change history

[July 19, 2018] Corresponds to NCSDK v2.05.00.02 / OpenCV 3.4.2 / FPS View
[July 27, 2018] Add an unnecessary version of OpenGL (USBCamera / MultiStick / MultiProcessing / RealSense)

Image of motion

MobileNet-SSD + Neural Compute Stick + RaspberryPi3 / MultiStick(3 Stick / Hard Motion)

Youtube: https://youtu.be/sQnFbRSqIA8

MobileNet-SSD + Neural Compute Stick(1 Stick) + RaspberryPi3 + USB Camera(640x480)
Youtube: https://youtu.be/_Cbt0gI8niQ

Environment

・RaspberryPi 3 + Raspbian Stretch

・NCSDK v2.08.01 (It does not work with NCSDK v1 v1 version is here)

・Intel Movidius Neural Compute Stick 1 piece

・OpenCV 3.4.2

・OpenGL

・numpy

・(UVC)USB-Web Camera

Building environment

Temporary extension of SWAP area

$ sudo nano /etc/dphys-swapfile
CONF_SWAPSIZE=2048

$ sudo /etc/init.d/dphys-swapfile restart;swapon -s

Installing packages

$ sudo apt-get update
$ sudo apt-get upgrade
$ sudo apt-get install python3-pip python3-numpy git cmake libqtgui4 libqt4-test

Installing NCSDK

$ cd ~
$ git clone -b ncsdk2 https://github.com/movidius/ncsdk.git
$ cd ncsdk
$ make install

Installation of OpenCV

$ cd ~
$ wget https://github.com/PINTO0309/OpenCVonARMv7/raw/master/libopencv3_3.4.2-20180709.1_armhf.deb
$ sudo apt install -y ./libopencv3_3.4.2-20180709.1_armhf.deb
$ sudo ldconfig

Installing OpenGL

$ sudo apt-get install python-opengl
$ sudo -H pip3 install pyopengl
$ sudo -H pip3 install pyopengl_accelerate
$ sudo raspi-config

「7.Advanced Options」-「A7 GL Driver」-「G2 GL (Fake KMS)」 and Activate Raspberry Pi's OpenGL Driver
Reboot

$ sudo reboot

Download complete set of resources

$ cd ~
$ git clone https://github.com/PINTO0309/MobileNet-SSD.git

Connect USB-WEB camera (UVC compatible) and Neural Compute Stick to RaspberryPi's USB port (self power USB-Hub required due to insufficient voltage when using Neural Compute Stick in multiple)
Connect RaspberryPi and display with HDMI cable
Running MobileNet-SSD

$ cd MobileNet-SSD
$ python3 MultiStickSSD.py

Reducing the SWAP area

$ sudo nano /etc/dphys-swapfile
CONF_SWAPSIZE=100

$ sudo /etc/init.d/dphys-swapfile restart;swapon -s

Procedure for generating original learning data

https://github.com/movidius/ncappzoo/tree/master/caffe/SSD_MobileNet
https://github.com/FreeApe/VGG-or-MobileNet-SSD
https://github.com/chuanqi305/MobileNet-SSD
https://github.com/avBuffer/MobilenetSSD_caffe

[Japanese] MobileNet-SSD

YoloV2 より超速 MobileNetSSD(MobileNetSSD)+Neural Compute Stick(NCS)+Raspberry Piによる爆速・高精度の複数動体検知
映像再生と物体検出は非同期実行。
マルチスティックを実現するために、マルチスレッドかつ OpenGL で実装している。
【Qiita記事】 https://qiita.com/PINTO/items/b97b3334ed452cb555e2
【RealSense D435 + MultiProcessing対応版】 https://github.com/PINTO0309/MobileNet-SSD-RealSense.git

変更履歴

[2018/07/19] NCSDK v2.05.00.02対応 / OpenCV 3.4.2対応 / FPS View対応

動作イメージ

MobileNet-SSD + Neural Compute Stick + RaspberryPi3 / MultiStick(3本/Hard)