MobileNet SSD model for Motion software

MobileNet SSD model here is intended for use with Intel's Movidius Neural Compute Stick hardware, Intel's OpenVINO framework, and Motion software.

Upstream Motion currently only supports pixel change detection algorithm. To use motion with a different detection algorithm like neural network based detection, you could try my alt_detection motion branch that supports alternate detection plugin. This alternate detection plugin is a generic interface that allows users to use any detection library they want. If you wish to use my MobileNetSSD detection library, head to my lib_openvino_ssd project.

Pre-trained MobileNet SSD files

I downloaded the pre-trained MobileNet SSD files from https://github.com/chuanqi305/MobileNet-SSD and converted them into OpenVINO supported formats (bin and xml files). Note that only full version of OpenVINO comes with model optimizer that does the model conversion.

• Caffe model was from here: https://drive.google.com/open?id=0B3gersZ2cHIxRm5PMWRoTkdHdHc
• Prototxt file was from here: https://raw.githubusercontent.com/chuanqi305/MobileNet-SSD/923b3128f25262b5010cef67e4fb9e4b6728ae7b/voc/MobileNetSSD_deploy.prototxt

To generate the MobileNet SSD graph file, run:

python3 /opt/intel/openvino/deployment_tools/model_optimizer/mo.py --input_model MobileNetSSD_deploy.caffemodel --data_type FP16 --output_dir . --mean_values [127.5,127.5,127.5] --scale_values [127.5]

MobileNetSSD expects the input BGR values from 0 to 1, so we add --mean_values [127.5,127.5,127.5] --scale_values [127.5] to scale it to 0 - 255.

How to install and run motion software with MobileNetSSD alternate detection library

1. Install OpenVINO raspbian release (2019-R3) on your Raspberry Pi. Follow the instructions. Alternatively, you could download the OpenVINO raspbian release here and unpack into /opt/intel/openvino directory.
   • sudo mkdir -p /opt/intel/openvino
   • sudo tar -xf l_openvino_toolkit_runtime_raspbian_p_<version>.tgz --strip 1 -C /opt/intel/openvino
2. Git clone the lib_openvino_ssd library.
   • git clone https://github.com/jasaw/lib_openvino_ssd
3. Build the lib_openvino_ssd library.
   • Install dependencies. sudo apt-get install libjpeg libavutil-dev libswscale-dev
   • cd lib_openvino_ssd
   • make -j4
4. Test the lib_openvino_ssd library.
   • cd openvino_ssd_test
   • make -j4
   • Copy a few jpg files with people in the images.
   • Edit ../libopenvino.conf to make sure the MODEL_BIN and MODEL_XML point to the MobileNetSSD_deploy.bin and MobileNetSSD_deploy.xml respectively.
   • Test the SSD library by running ./ssd_test -l ../libopenvinossd.so -c ../libopenvino.conf photo_1.jpg photo_2.jpg. Make sure your NCS stick is connected.
   • If the test is successful, it will output png files with the detection result drawn on the png image.
5. Git clone my motion alt_detection motion branch.
   • git clone -b alt_detection https://github.com/jasaw/motion.git
6. Build and install motion.
   • Install dependencies. sudo apt-get install autoconf automake build-essential pkgconf libtool libzip-dev libjpeg-dev git libavformat-dev libavcodec-dev libavutil-dev libswscale-dev libavdevice-dev libwebp-dev gettext libmicrohttpd-dev
   • cd motion
   • autoreconf -fiv
   • ./configure
   • make -j4
   • sudo make install
7. Specify which alternate detection library to load by adding the below lines in motion.conf file.
   • alt_detection_library /home/pi/lib_openvino_ssd/libopenvinossd.so
   • alt_detection_conf_file /home/pi/lib_openvino_ssd/libopenvino.conf
8. Specify which camera to use alternate detection by adding the below lines in the camera config file, e.g. camera-1.conf file.
   • alt_detection_enable on
   • alt_detection_threshold 75