Skeletons Detection Service

Streams

Name	Input (Topic/Message)	Output (Topic/Message)	Description
Skeletons.Detection	CameraGateway.\d+.Frame Image	SkeletonsDetector.\d+.Detection ObjectAnnotations	Detect skeletons on images published by cameras and publishes an ObjectAnnotations message containing all the skeletons detected
Skeletons.Detection	CameraGateway.\d+.Frame Image	SkeletonsDetector.\d+.Rendered Image	After detection, skeletons are drew on input image and published for visualization. NOTE: This stream will be deprecated after mjpeg server became able to render any ObjectAnnotations

• Note: run the is-skeletons-detector-stream to use this function.

RPCs

Service	Request	Reply	Description
SkeletonsDetector.Detect	Image	ObjectAnnotations	Same purpose of stream shown above, but offered with a RPC server.

• Note: run the is-skeletons-detector-rpc to use this function.

About

This detector uses an implementation based on OpenPose method. You can choose between two available models, CMU or MOBILENET_THIN, both are based on COCO Model that contains 18 joints. These joints are identified by a subset of an enumeration message called HumanKeypoints. The joints of COCO model are: NOSE, NECK, RIGHT_SHOULDER, RIGHT_ELBOW, RIGHT_WRIST, LEFT_SHOULDER, LEFT_ELBOW, LEFT_WRIST, RIGHT_HIP, RIGHT_KNEE, RIGHT_ANKLE, LEFT_HIP, LEFT_KNEE, LEFT_ANKLE, RIGHT_EYE, LEFT_EYE, RIGHT_EAR and LEFT_EAR.

Beyond the model, you can choose the CNN (Convolutional Neural Network) input size changing width and height parameters of resize field on options.json file. Both must be multiples of 16. The CNN output layer can also be resize, for that change the parameter resize_out_ratio. Both parameters impact on detection time and quality.

Developing

Once it's necessary to install many dependencies including NVIDIA Cuda libraries, a Docker image is created before the deployment image. This image can be user to perform tests during development. For that, simply run the script etc/docker/pack.sh:

cd etc/docker/
bash pack.sh

If it is the first time you run this script, a message will be prompt indicating that the image is-skeletons-detector/dev wasn't found, then the build process will occur. Now just run the command below inside the project directory.

docker run -ti --rm --runtime=nvidia --network=host -v `pwd`:/devel is-skeletons-detector/dev bash

In case you need to make any change on options protobuf file, will be necessary to rebuild the python file related to it. For do that, simply run the script src/conf/gen_pb_py.sh.

cd src/conf/
bash gen_pb_py.sh