Shulou(Shulou.com)06/02 Report--

This article is about how to use the webcam and OpenCV in Python to build a motion detector. The editor thinks it is very practical, so I share it with you. I hope you can get something after reading this article.

Motion detection refers to the detection of whether the position of the object relative to the surrounding environment has changed. Next, let's implement a motion detector application using Python.

The motion detector can accomplish the following tasks:

1) look up the time in front of the screen when working from home

2) monitor the child's time in front of the screen

3) trespassing was found in your backyard

4) find unwanted public / animal activities around your room / house / alley.

To implement the motion detector program, we need to have the following conditions:

1) hardware requirements: a computer with a network camera or any type of camera.

2) Software requirements: Pyhton3 or higher.

3) additional requirements: have some interest in motion detection.

Next we will complete the construction of the application step by step.

First, we will capture the first frame through the webcam and treat it as the base frame, as shown in the following figure. Motion is detected by calculating the phase difference between the object in the reference frame and the new frame object. We also call the result a Delta frame.

Next, we will use pixel intensity to optimize Delta frames, which are called threshold frames. In addition, we will apply some complex image processing techniques, such as shadow elimination, expanding contours, etc., to extract the object on the threshold frame. Here is what you want to achieve:

Detected object

When the object enters the frame and exits the frame, we can easily capture the timestamps of the two frames. Therefore, we will be able to accurately find the relevant clips in the video.

We hope that all the partners can implement this program on their own, so we don't embed the code directly.

Starting with the most basic installation, we need to install Python3 or later, and use pip to install the pandas and OpenCV libraries. When all this work is done, our preparatory work will be completed.

Step 1: import the required libraries:

Step 2: initialize variables, list, data frame:

In the following code, we will learn when to use each of the items mentioned above.

Step 3: capture video frames using a webcam:

There is a built-in function in OpenCV that can turn on the camera and capture video frames. Where the input parameter "0" represents the camera whose computer hardware port number is 0. If we have multiple cameras or closed-circuit television settings, we can provide the corresponding port number through this parameter.

Step 4: convert the captured frame into a grayscale image and apply Gaussian blur to remove noise:

Because each pixel in the color picture has three color channels, in fact, we do not need to use so much information, so we first convert the color frame into grayscale frame. Then Gaussian blur is used to smooth the image to improve the detection accuracy. In the Gaussian ambiguity function, we use the second parameter to define the width and height of the Gaussian kernel, and the third parameter to define the standard deviation. Here we can use a standard value with a core size of (21) and a standard deviation of 0. To learn more about Gaussian smoothing, please refer to:

Smoothing Images-OpenCV 2.4.13.7 documentation In an analogous way as the Gaussian filter, the bilateral filter also considers the neighboring pixels with weights...

Docs.opencv.org

Step 5: capture the first grayscale frame

The first frame is the reference frame for the whole process. Motion is detected by calculating the phase difference of a specific object between the reference frame and the new frame. When shooting the first frame, there should be no movement in front of the camera for a particular object. But the resulting first frame does not require subsequent processing, so we can skip the subsequent process with the continue statement.

Step 6: create Delta frames and threshold Fram

Now, we need to find out the difference between the first frame and the current frame. Therefore, we use the absdiff function and call the result a delta frame. It is not enough to find a difference for our use case, so we need to define a pixel threshold that can be seen as a real object.

We can choose 30 pixels as the standard threshold and define the color of the standard threshold as white (color code: 255). The binary threshold function THRESH_BINARY returns a tuple value where only the second term ([0] is the first term and [1] is the second term) contains the generated threshold frame. The binary threshold function is used to deal with discontinuous functions with two discrete values, such as 0 or 1. If there is no object in front of the camera, we treat the state of the current frame as 0; if there is an object in front of the camera, we treat the state of the current frame as 1.

For more information about threshold image processing, please refer to:

Miscellaneous Image Transformations-OpenCV 2.4.13.7 documentation Performs a marker-based image segmentation using the watershed algorithm. The function implements one of the variants...

Docs.opencv.org

Step 7: inflate the threshold frame and find outline pixels in it

"our eyes are always attracted by light, but there is more in the shadows." -Gregory Maguire

Each part of the object leaves a shadow on the background or other parts of itself. It always seems to confuse us. For example, the shadow cast by the nose on the lips, or the shadow cast by a larger still object on a small object next to it. The floating light source, multiple light sources with different luminous intensity, the curtains of your room, the direction and viewing angle of the light source, and so on will have a certain impact on the shadow.

Here are some of the interference found in real-time captured frames. Therefore, in order to minimize the noise, we need to filter the image. In the inflation function Dilate, we can set the smoothness by setting the number of iterations. The more iterations, the higher the smoothness, and the longer the processing time. Therefore, it is recommended that you keep the standardization setting to 3. The "None" parameter in the dilation function indicates that we do not need an element structure in our application.

For more information about inflation, you can refer to:

Image Filtering-OpenCV 2.4.13.7 documentation Functions and classes described in this section are used to perform various linear or non-linear filtering operations...

Docs.opencv.org

After filtering, we need to find the outline of the object in this frame. We use the outline in the current frame to identify the size and position of the object. To achieve this, we pass a copy of the frame to the findCounters method and use this copy to find the outline. The reason for using the copy is that we do not want the contour recognition to affect the original filtered frame.

There is a hassle here because we have to store the profile in a tuple and only need to use the first value of that tuple. See the syntax for declaring tuples in Python3: (name,_).

Now, we just need to find the outer outline of the object on the filter layer. For our use case, contours other than extreme external contours are useless. Therefore, we must use some approximate methods to optimize the contour extraction process. For example, if you use curve approximation or curve interpolation, you can also use simple chain approximation rules, that is, to compress horizontal, vertical, and diagonal segments, leaving only their endpoints. Therefore, we can quickly get the best fitting profile.

Step 8: find the outline area and form an endpoint in the rectangle:

In fact, we don't want to catch small objects like insects, but we want to catch large objects like people or animals. Therefore, we adopt the concept of contour region, that is, skip objects with an area of less than 10000 pixels. For contours larger than this area, we set the state to 1, that is, objects are detected.

For more information about contours in image processing, please refer to:

Structural Analysis and Shape Descriptors-OpenCV 2.4.13.7 documentation Draws contours outlines or filled contours. The function draws contour outlines in the image if or fills the area...

Docs.opencv.org

Now we use the boundingRect function to capture the coordinates of the outline. Then we use these coordinates to draw a rectangle of a specific color and thickness on the color frame. This rectangle describes the actual detected object.

Step 9: capture the timestamp of the object when it enters and exits the frame (scene)

The status list status_list stores a value of 0: indicates that the object is not detected, and 1: indicates that the object is detected. The moment the state value changes from 0 to 1 is the moment the object enters the frame. Similarly, the moment when the state value changes from 1 to 0 is the moment when the object disappears from the frame. Therefore, we can get the timestamps of these two switching events from the last two values of the status list.

Step 10: display all different pictures (frames)

Using the imshow () method, we will display each frame in a separate window and compare it.

We use the waitKey function to delay the process until a key is pressed. Here, we use waitKey (1) to get continuous real-time feedback from the camera. When you want to stop shooting a video, just press the Q key on the keyboard.

We also need to capture the last timestamp while pressing "Q", as this will help the program end the process of capturing video from the camera and generate time data.

The following is the actual image output generated using the application. The first image represents the four frame types of the base frame, and the second image represents the four types of frames with the object. Can you compare the difference?

Baseline First Frame

Frame with a detected object

Step 11: generate time data

So far, all timestamps have been stored in the data-frame variable of pandas. To get more information from the generated data, we will export the data-frame variable to the csv file on our local disk.

Don't forget to release the video variable, because it takes up a lot of space in memory. Destroy all windows at the same time to avoid unnecessary errors

This is what the generated csv looks like. As we can see, this object has been detected three times before the end of the program. You can view the start and end times and calculate the time the object is in front of the camera.

Isn't this app exciting enough? Is this application far away from typical boring programming? Internet of things enthusiasts can even deploy this program to the raspberry pie server Raspberry Pi and work miracles!

This is how to use the webcam and OpenCV in Python to build motion detectors. The editor believes that there are some knowledge points that we may see or use in our daily work. I hope you can learn more from this article. For more details, please follow the industry information channel.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.