Opencv Number Detection How To Get Live

Opencv Number Detection How To Get Live

In our previous Raspberry Pi CCTV monitoring article, we learned how to get live CCTV video from a DVR using RTSP and display on a Raspberry Pi, do check that out before proceeding.

In this article, we will learn how to leverage the power of OpenCV and build a Raspberry Pi Motion Detection system on our live CCTV footages.

If you do not have a CCTV installed, you can still build a Raspberry Pi Surveillance system by connecting USB cameras directly to your Pi.

And if you are not a big fan of Pi and Python, you can build something similar with ESP32, refer to ESP32 Wi-Fi Door Bell for more details.

If an activity is detected on any camera, our Raspberry Pi will automatically change to that particular camera screen and highlight which activity took place, all this in real-time with just 1.5 seconds lag.

I have also added an alarm feature, like a buzzer to which can alert the user by beeping if an activity is detected.

But you can easily scale this up to send a message or E-mail or what not Exciting right Lets get started.

You can either follow the above tutorial to compile your OpenCV which will take hours but is more reliable for heavy projects or just install it directly from pip using the following commands.

After Interfacing 5-Inch Display with Raspberry Pi, we can directly mount the buzzer to the backside of the Display which has extended some GPIO pins for us.

As you can see among the extended pins, most of the pins are utilized by the Display itself for a touchscreen interface.

But still, we have pins 3,5,7,8,10,11,12,13,15,16, and 24 that have no connection and we can use it for our own application.

Initially, I tried switching between all four cameras to look for motion but it was very laggy (about 10 seconds).

So I combined all four cameras to a single image and did all the motion detecting activities on that image.

I wrote two functions, namely, create a camera and read camera.

Note that RTSP URL ends with 02, which means I am using the sub-stream video feed which will be low in resolution and hence faster to read.

Also, the type of video codec you use also contributes to speed, I experimented with different codes and found FFMPEG to be the fasted of all.

Once this main screen is ready, we will do all our OpenCV work on this image.

Inside the while loop, we start by reading two different frames namely, frame1 and frame2, then convert them to grayscale.

It is also common to blur and dilate the image to avoid sharp edges.

If the area is bigger than a specified value in the variable motiondetected, then we consider that as an activity and draw a box around the change to highlight it to the user.

We can find that since we know the x and y values of the motion.

We compare these x and y values to the location of each screen to find which screen gave activity and we again crop that particular screen, so that we can display it on the pi touch screen.

The if statement checks if the motion was detected in screen 3 and increments a variable called trigalarm.

You can detect any screen of your choice or even on multiple screens.

The reason for this count is that sometimes I noticed shadows or birds created fake alarm.