Protect Your Garden Against Cats

1. Story

Almost one year ago we moved from an apartment to a house together with our cat and baby daughter. The nightmare has started when our cat started using the garden as a toilet.

There was several tries to push him to come back to the cat litter, but all of them didn’t succeed, he really liked the garden as much as we do. So I started thinking about something which could prevent the cat to go there, initially I thought about ultrasonic speaker or compressed air to scare him. But both are not good for open spaces.

Then I thought about an presence sensor spraying water, but I did not want my baby daughter to get wet instead of the cat, so this problem required a more sophisticated approach. A computer vision system which could differentiate between a person and a cat, interpreting images from a live camera and triggering a Solenoid valve which can sprinkle water in the naughty cat.

2. To have more fun

Ive added a video recording feature, in order to monitor the cat behavior. The cat tried several different strategies before giving up, first he found a blind spot which forced me to reposition the camera, then he noticed that during the night it was harder to be detected, so I had to install a garage light activated by motion. Don't underestimate cats!

3. Assembly

The Project has few components, I used a plastic box fixed in the wall close to the water tap. My USB camera have a long cable in order to allow to try different possitions. The initial position I selected was too overshadowed generating false recognitions, and the camera was too high in the wall making the detection more difficult, the best position was 50 cm from the ground.

Please take in consideration:

• Water and electricity don´t get along well
• Raspberry overhearing
• Use a relay board to isolate the Raspberry port

I recommend isolate all hose connections from the electronics, I've assembled mine putting the solenoid at the bottom of the box, sealing the connectors with hot glue.

I also recommend use a fan to keep ARM processor cool, mine processing rate is around 70%, this level without fan can lead to overheating. Raspberry has a protection which reduces clock to avoid the processor to burn, but this can affect your system recognition speed and reduces life time of your components.

At last but not least, use a relay board that protects the raspberry againt the coil flutuation otherwise it will fry your port. You need a diode and a transistor arrangement to be safe.

The watering system connection/design is beyond the scope of this Project, therefore I will stick to the electronic and software parts.

4. Software

The Raspberry preparation is the most difficult part, the best guide I've seen is comming from www.pyimagesearch.com/2017/09/18/real-time-object-detection-with-deep-learning-and-opencv/

Which also provided a very comprehensive object detection concept explanation, by the way this was what I used as a base for my project. Many thanks to Dr. Rosebrock!!.

The algorithm is a quite simple, everything is build around the inference loop, which takes a image from video stream and passes it thru the deep neutral network in openCV.

Luckily this task doesn't require too much computer power considering an light network architecture as Mobilenet, so a Raspberry is enough for the required FPS, there are several pre trained models available in the internet.

I utilized a SSD Mobilenet trained in Pascal VOC dataset, which recognizes 20 classes including persons and cats what is perfect. Anyhow you can use another trained network without significative changes in the code.

My system performs about.5 FPS what is enough for the task but not a good rate if you need to monitor quick changes, but again this was enough for the purpose of the project.

Complete source code in GitHub.

5. Further improvements to enhance performance

• Split interference in a different threat
• Utilize a faster SD card
• Raspberry 4
• Use another network architecture as faster R-CNN, Yolo