A Node-RED node for Object Detection using TensorFlowJS CoCo SSD.
NOTE: The Tensorflow.js library will be installed automatically. However Tensorflow.js is only available on certain OS/Hardware/processor combinations. Therfore it might not automatically work on all platforms, if you are unlucky...
Either use the Node-RED Menu - Manage Palette option, or run the following command in your Node-RED user directory - typically ~/.node-red
npm i node-red-contrib-tfjs-coco-ssd
On a Pi - you will also need to run this command after the main install.
cd ~/.node-red
npm rebuild @tensorflow/tfjs-node --build-from-source
This node runs the CoCo Single Shot object detector on a jpeg image, delivered via an msg.payload in one of the following formats:
- As a string, that represents a file path to a jpg file.
- As a buffer of a jpg.
- As an https url that returns a jpg.
- As an html data:image/jpeg;base64, string
The CoCo-ssd model is loaded locally so it should work offline.
The model is currently trained to recognize the following object classes in an image:
person, bicycle, car, motorcycle, airplane, bus, train, truck, traffic light, fire hydrant, stop sign, parking meter, bird, cat, dog, horse, sheep, cow, elephant, bear, zebra, giraffe, backpack, umbrella, handbag, tie, suitcase, frisbee, skis, snowboard, sports ball, kite, baseball glove, skateboard, surfboard, tennis racket, bottle, wine glass, cup, fork, knife, spoon, bowl, banana, apple, sandwhich, orange, broccoli, carrot, pizza, donut, cake, chair, couch, potted plant, bed, dining table, toilet, tv, laptop, mouse, remote,, keyboard, cell phone, microwave, oven, toaster, sink, refrigerator, book, clock, vase, scissor, teddy bear, hair drier, toothbrush
The following example will demonstrate how easy it is to recognize objects in images.
TIP: avoid adding too many CoCo-ssd nodes in a flow. This avoids long startup times (due to loading the same model N times). From a performance point of view, it is better to reuse a single coco-ssd node for multiple sources if at all possible.
Some of the supported object types (like cars, persons, ...) are very useful in a typical IOT environment, e.g. to implement a video surveillance application in Node-RED.
The following example flow shows how to recognize these object classes in an input image. Note that this flow requires that the node-red-contrib-image-output is installed, to be able to display the analyzed jpg images!
The recognition results will be displayed in the debug node's status:
[{"id":"9798f509.db7bc8","type":"inject","z":"eeee575.b4b80a8","name":"Group of people","repeat":"","crontab":"","once":false,"onceDelay":0.1,"topic":"","payload":"","payloadType":"date","x":220,"y":780,"wires":[["d6d7a98f.929108"]]},{"id":"d6d7a98f.929108","type":"http request","z":"eeee575.b4b80a8","name":"","method":"GET","ret":"bin","paytoqs":false,"url":"https://upload.wikimedia.org/wikipedia/commons/b/b3/Team_Queerala.jpg","tls":"","persist":false,"proxy":"","authType":"","x":450,"y":780,"wires":[["3e623d1a.c9fd42"]]},{"id":"3e623d1a.c9fd42","type":"tensorflowCoco","z":"eeee575.b4b80a8","name":"","modelUrl":"http://localhost:1880/coco/model.json","scoreThreshold":0.5,"passthru":"bbox","x":650,"y":780,"wires":[["14fa8ce2.fc7043","4ec7b21f.c2d45c"]]},{"id":"fa56b166.50b0d","type":"http request","z":"eeee575.b4b80a8","name":"","method":"GET","ret":"bin","paytoqs":false,"url":"https://upload.wikimedia.org/wikipedia/commons/9/9d/Pedestrian_checking_before_crossing_the_road.jpg","tls":"","persist":false,"proxy":"","authType":"","x":450,"y":840,"wires":[["3e623d1a.c9fd42"]]},{"id":"bc6d8978.d30338","type":"http request","z":"eeee575.b4b80a8","name":"","method":"GET","ret":"bin","paytoqs":false,"url":"https://upload.wikimedia.org/wikipedia/commons/c/cb/Old-style_VAZ_car_in_Kolpino_with_USSR-time_car_number.jpg","tls":"","persist":false,"proxy":"","authType":"","x":450,"y":900,"wires":[["3e623d1a.c9fd42"]]},{"id":"17e6db88.6c8274","type":"http request","z":"eeee575.b4b80a8","name":"","method":"GET","ret":"bin","paytoqs":false,"url":"https://upload.wikimedia.org/wikipedia/commons/3/36/Movement_and_cars.jpg","tls":"","persist":false,"proxy":"","authType":"","x":450,"y":960,"wires":[["3e623d1a.c9fd42"]]},{"id":"2666c156.8217fe","type":"http request","z":"eeee575.b4b80a8","name":"","method":"GET","ret":"bin","paytoqs":false,"url":"https://upload.wikimedia.org/wikipedia/commons/3/3f/Pedestrian_crossing_street.jpg","tls":"","persist":false,"proxy":"","authType":"","x":450,"y":1020,"wires":[["3e623d1a.c9fd42"]]},{"id":"14fa8ce2.fc7043","type":"debug","z":"eeee575.b4b80a8","name":"","active":true,"tosidebar":true,"console":false,"tostatus":true,"complete":"classes","targetType":"msg","x":870,"y":780,"wires":[]},{"id":"4ec7b21f.c2d45c","type":"image","z":"eeee575.b4b80a8","name":"","width":"250","data":"image","dataType":"msg","thumbnail":false,"active":true,"outputs":0,"x":880,"y":860,"wires":[]},{"id":"2b3074e9.e13fbc","type":"inject","z":"eeee575.b4b80a8","name":"Cars and persons","repeat":"","crontab":"","once":false,"onceDelay":0.1,"topic":"","payload":"","payloadType":"date","x":230,"y":840,"wires":[["fa56b166.50b0d"]]},{"id":"fa7c8056.39de2","type":"inject","z":"eeee575.b4b80a8","name":"Single car","repeat":"","crontab":"","once":false,"onceDelay":0.1,"topic":"","payload":"","payloadType":"date","x":200,"y":900,"wires":[["bc6d8978.d30338"]]},{"id":"68490c5d.480394","type":"inject","z":"eeee575.b4b80a8","name":"Multiple cars","repeat":"","crontab":"","once":false,"onceDelay":0.1,"topic":"","payload":"","payloadType":"date","x":210,"y":960,"wires":[["17e6db88.6c8274"]]},{"id":"9b5463ca.5bc32","type":"inject","z":"eeee575.b4b80a8","name":"Pedestrians","repeat":"","crontab":"","once":false,"onceDelay":0.1,"topic":"","payload":"","payloadType":"date","x":210,"y":1020,"wires":[["2666c156.8217fe"]]}]
All images used are freely-licensed offered by Wikimedia Commons.