Wolf and herd protection through Artificial Intelligence

Challenge

The wolf is becoming an increasingly present threat to herd animals , because more and more often there are attacks in Germany, Austria and Switzerland. Currently, it is not possible to ensure the protection of the herd animals as well as the wolf at the same time.

Solution

The goal was to use neural networks to enable animal owners to actively protect themselves from wolf attacks. This is to be provided in the form of a user-friendly app that is connected through an API to our in-house developed artificial intelligence. This AI will be used primarily to detect wolves in camera footage.

1. Potential analysis

Through intensive exchange with the customer, we gained an overview of their ideas and requirements for the image recognition software and app. We captured these insights in a click dummy app. Thus, we were able to define tools, technologies and architectures in coordination with them in order to work out the best possible potential from this process.

2. Use case definition

Based on the process analysis, we defined the use case of the desired app. Such a use case contains all possible scenarios that can be performed with the help of the system.

Name

Wolfswarner

Brief description

The task is to provide a trained AI in an app to immediately detect and warn users about wolves and humans based on video material.

Actors

Masasana GmbH

Oliver Kubitz

Precondition

The AI is trained. In addition, the app is developed and installed at the user’s site. The cameras are set up and the RTSP link of these is publicly available.

Postcondition

The user is aware of the presence of a wolf/human on their property.

Process – normal scenario

1. The user authenticates themself in the app.

2. The user integrates their own camera(s) into the app via a subscription model.

3. The backend receives live data from each integrated camera.

4. Artificial intelligence reviews and processes the video data.

5. AI-assisted detection software detects a wolf and/or human.

6. The system issues an automatic live alert to the user, who receives it via the app.

Process – alternative scenario

2.1 The user does not have a subscription.

2.1.1 The user purchases a subscription of their choice via the app.

2.1.2 Continue as in step 2.

5.1 The detection software does not detect a wolf/human.

5.1.1 The system does not issue a message to the user.

Process – error scenario

1.1 The user cannot authenticate in the system.

1.1.1 The user registers in the app.

1.1.2 Continue as in step 2.

2.2 The camera does not connect to the system.

2.2.1 The system issues an error message.

3.1 The system does not receive video data.

3.1.1 The system issues an error message.

3. Iterative model creation

In order to get an overview of the architecture and content of the app, we developed a mockup together with our customer in a workshop. One of the results of this workshop was a click dummy of the app. Based on this dummy and mockup, our team developed a complete app in which the artificial intelligence is integrated via the backend.

Fig. 1: Mockup of the App

After the concept development, we started with the data acquisition. For the project, around 300,000 photos of wolves were collected, analyzed and labeled to train the artificial intelligence. Many of these came from the Anholter Schweiz biotope wildlife park in Isselburg. In addition, thousands of images of dogs, sheeps, horses, cows and humans were included so that the artificial intelligence can recognize the difference between the predators and harmless farm animals/humans. This should drastically reduce the number of false alarms. In the end, this artificial intelligence was able to recognize wolves, humans, dogs, sheep, horses and cows by day, night as well as in fog.

Fig. 2: Camera setup in app
Fig. 3: Push notification in case of wolf detection

In addition to the backend development, it was particularly important to make the frontend user-friendly. Accordingly, all important functions and information of the app should be recognizable and accessible at a single glance. This included, above all, the simple setup of the cameras. When the app is started for the first time, the intro screen displays a guide that explains how to integrate the cameras. The subscription model in the app allows the user to connect up to 10 cameras to the software. Almost any commercially available security camera can be connected to the app and only needs to support the RTSP protocol and be accessible via a public IP address or a DynDNS.

When setting up the camera, the user can choose to be informed only about wolves and/or humans (see Figure 2).

Furthermore, the app has been expanded to include a support area with a Frequently Asked Questions (FAQ) section, which provides immediate answers to common questions for the user.

In the app, the user can create a profile with a username, email address and profile picture. This data is stored in a backend database. The user’s password is stored in encrypted form in this database to guarantee data protection.

Fig. 4: Wolf detection alarm in app

When a warning is received via push message, the live image of the camera recording is displayed. This image can be scaled up using a zoom function. The artificial intelligence thereby indicates whether it is a wolf or a human. Through the image, it is possible for the user to check if the AI’s statement is True Positive or False Positive. If it is incorrectly not a wolf or human (false positive), a false alarm can be reported in the app.

4. Integration and result analysis

After the completion of the app, at the request of the customer, we published it in the App Store for IOS as well as in the Play Store. Thus, the app is available for the smartphone and tablets since June 2022. In addition, we are constantly enhancing the artificial intelligence in collaboration with our client to increase the recognition rate.

Media reports

WDR (1), WDR (2), O. Kubitz, MDR

 

 

 

Cookie Consent with Real Cookie Banner