/PPE-detection-deepstream

PPE detection of helmets(construction) using Nvidia Deepstream. Model trained using Nvidia TLT.

Primary LanguageC

PPE-detection

A Jetson based DeepStream application to detect,track and count people crossing with and without PPE at hazardous sites.

The application was compiled in ubuntu 18.04 system with Nvidia Cuda 10.2 + CuDNN 8.05 , TensorRT 7.0.0 and Deepstream 5.0. Tested on Jetson Nano and TX2 with Jetpack 4.4

Applications

This repo contains two applications for different use cases.

1. PPE-primary application

primary

  • When a person crosses a line following labels are detected

    1. helmet
    2. person_with_helmet
    3. person_without_helmet

  • At the instace of a person crossing a user-defined line/region, the person is counted as person_with helmet or person_without_helmet.

2. PPE-secondary application

secondary

  • This application is intented when we have data only for the PPE objects. Not all the time we will be able to get images of PPE with/without person.

  • It uses peoplnetv2 as primary inference engine and our custom model to detect helmets as secondary inference engine.

  • People are detected using peoplenetv2.

  • The detected region is passed to secondary inference engine for detecting helmets.

  • At the instace of a person crossing if the secondary inference detects helmet then its counted as person_with_helmet.

  • At the instace of a person crossing if the secondary inference does not detect helmet then its counted as person_without_helmet.

Deepstream Setup

Documentation to setup Jetson Device can be found here.

1. Install System Dependencies

sudo apt install \
libssl1.0.0 \
libgstreamer1.0-0 \
gstreamer1.0-tools \
gstreamer1.0-plugins-good \
gstreamer1.0-plugins-bad \
gstreamer1.0-plugins-ugly \
gstreamer1.0-libav \
libgstrtspserver-1.0-0 \
libgstrtspserver-1.0-dev \
libjansson4=2.11-1 \
libjson-glib-dev \
libgstreamer1.0-dev

2. Install Deepstream

Download Deepstream 5.0 and run sudo apt install ./<deepstream_deb_file>.deb

Run Application

Clone the repository. $ git clone https://github.com/Maouriyan/PPE-detection-deepstream.git $ cd PPE-detection-deepstream/ppe-application/ Make : One Makefile will make both applications , you can also make respective applications using the Makefile in respective folders. $ sudo make $ sudo make install Run the PPE-primary application. $ cd primary-app/ $ sudo ./ppe-primary -c ../../configs/ppe_primary.txt -t Run the PPE-secondary application. $ cd secondary-app/ $ sudo ./ppe-secondary -c ../../configs/ppe_secondary.txt -t

Note: If you are using the device headless , disable EGL sink at configs/ppe_primary.txt

[sink0]
enable=0
#Type - 1=FakeSink 2=EglSink 3=File
type=2
sync=1
source-id=0

[osd]
enable=0
gpu-id=0

Test Results

result

Training

  • To train the model , add more datasets use this.