tjf / visionalert

Monitor your security cameras for the presence of various objects (people, animals, etc) using Tensorflow

visionalert

Visionalert is a python application that monitors RTSP streams for objects and sends a push notification to your Android device when something is detected. Upon detecting a configured object, a notification is scheduled to be sent after 5 seconds containing the image containing the image with the highest confidence score detected since the alert was triggered.

Requirements

• Tensorflow Lite SSD MobileNet model and label file from here
• Gotify instance and Android application for push notifications
• MinIO for image hosting (This should work with Amazon S3 as well)

Installation

The easiest way to get this running is by using Docker.

Download the TensorFlow model from the link above, unzip it into a directory and create a configuration file named config.yml like the following:

1. Download the Tensorflow model and unzip it into your current directory. This also supports the Google Coral EdgeTPU Accelerator but you need to use a compatible SSD model with it. Note: Only configure the model, use the labelmap from the above link. This is due to a different format for the labelmap that comes with the SSD MobileNet for the EdgeTPU.

$ wget -q https://storage.googleapis.com/download.tensorflow.org/models/tflite/coco_ssd_mobilenet_v1_1.0_quant_2018_06_29.zip
    
$ unzip coco_ssd_mobilenet_v1_1.0_quant_2018_06_29.zip 
Archive:  coco_ssd_mobilenet_v1_1.0_quant_2018_06_29.zip
  inflating: detect.tflite           
  inflating: labelmap.txt            

2. Create a configuration file named config.yml in your current directory. There is a sample configuration file located here

3. Start the visionalert service by executing the following command. Note, this will launch the container into the background.

docker run -d -v ${PWD}:/conf --name visionalert tylerfrederick/visionalert:latest

4. You can monitor the behavior of the application or view the logs by running docker logs visionalert

Roadmap

• Object area boundaries. Suppress alerts if a detected object is too large or small to help prevent false alerts.
• Other notification strategies. Considering MQTT so I can use it to trigger Node-Red flows.

Contributing

Pull requests are welcome!