Introduction

dialEye is an application for reading the rotating dials of a utility meter (water, gas, etc.). It uses image processing to determine the angles of the rotating dial needles and combines the result to a counter value. dialEye needs to be configured and clibrated so that it is aware where in the meter image the dials are located and what are the 0-angles of the dial needles. In addition special positioning areas can be configured to deal with possible movement of the camera in relation to the meter.

The meter counter value represents the position the dial needles are at a time. Periodically metering and comparing the counter value change, the consumption value can be calculated.

This howto is based on an example provided with the dialEye application package (sample dialEye.conf configuration file and sample calibration image docs/dialeye_calibration.jpg).

Requirements

Requirements for running the application are:

• Python series 2 interpreter 2.4 or newer (Python version 3 not supported) (www.python.org)
• Python Imaging Library (www.pythonware.com/products/pil/)

Hardware and camera

dialEye is not dependent on certain image size, camera type or image format. However there are some rules that should be taken into account when planning the camera setup. You might use a network capable camera or a $0.99 web camera from Hong Kong (I have done both with success, though the cheap webcam one did stop working after year of successful operation). As you see from the example images below, the picture does not have to be crystal sharp. There is one key driving element though:

Some tips to consider when planning your setup:

• Mount the camera so that it is more likely that the meter will stay in the same position in the camera image and that no-one kicks or moves the camera.
• Keep the light conditions as constant as possible. I use external white LED light to light the meter.
• Detection algorithm works better with full color picture. Do not use IR-light because it flattens the colors and "discards" color information making it harder to detect the dials.
• When you light the meter, be sure there are no reflections on the meter glass on top of the dials that are being detected. Light reflections are OK in areas outside the dial circles.
• If sunlight/outside light reaches your meter, consider covering the setup to make things constant.

dialEye is able to get the image from network camera using HTTP (Basic auth is supported) or from an image file on the file system.

Example photos of existing camera setups:

• Installation 1: D-Link network camera (WLAN) mounted to the meter stand with all the construction iron I could find... and covered with sheet metal to protect from condence water dripping from above equipment.
• Installation 2: Cheap webcam mounted over the meter using a cookie can. Cam LEDs have been changed from IR to white light. Cookie can protects from sunlight.

Installing software

dialEye is simply installed by unpacking the distribution package to suitable location in the target system directory tree.

Configuration

The basic dial and needle value detection requires information about where in the meter image the dials are located, how big is the dial area and what is the geometry of the needle. Also detection treshold value can be adjusted. In addition the application can be configured to first search known parts of the supposed image to properly locate the dials should the positioning of the meter and camera have shifted slightly from some reason.

dialEye package comes with an example configuration file dialEye.conf that matches the example calibration image found in the docs folder. The following prodcedure explains how the sample configuration has been done and meaning of each value. The exmaple configuration file also includes comment rows (beginning with #-mark) that further document possible configuration parameters and their possible values.

Calibration image

First thing to configure is to obtain a calibration image from your meter and camera combination. Use an image that is taken with your installed setup, camera position and light conditions. Below is the calibration image on this example - the starting point of my configuration.

The calibration image is configured in the configuration file with key CALIBRATION_IMAGE:

    CALIBRATION_IMAGE = docs/dialeye_calibration.jpg

Dials

Configuration first lists the dials on the calibration image with @DIAL configuration keys. The image may contain from 1 to unlimited amount of dials. The dial configuration lists dials in significance order. The most significant dial must be configured as topmost in the configuration file. The dials will be processed in the order they are listed and will produce the output value digits in that order.

The syntax of the @DIAL-row is:

    @DIAL=center_x:center_y:meter_radius:inner_radius:needle_angle:needle_angle2:zero_angle:rot_dir[:needle_color]

Example @DIAL-row (without needle color):

    @DIAL = 402:253:33:16:-1.0:50.0:3.5:0

Example @DIAL-row (with needle color):

    @DIAL = 402:253:33:16:-1.0:50.0:3.5:0:152:57:65

Let's start with most significant dial and by guessing the center point of it (center_x, center_y) upper left corner being the (0,0) point. Then set the meter_radius for example to 20 and left of the values to zero:

    @DIAL = 350:250:20:0:0:0:0:0

dialEye can be asked to visualize the configuration on top of the calibration image. The visualized configuration is stored to a image file dialeye_conf.png or shown in a new window if -g option is given and the system has graphical environment available (Windows environment or X). The commands used to view the configuration:

    python dialEye.py showconf

    python dialEye.py -g showconf

The outputted image (in file or screen) will show the meter outer diameter with green and at this point a red thin needle ending in the configured center point.

Correct the center point and dial outer diameter so that the circle is exaclty in the middle of the dial and the circle area covers fully the area where the needle rotates. An image editing application (eg. GIMP or Paint.NET) may be helpful to find out the pixel locations in the image, but trial error is also feasible way to proceed. After the values are correct, the showconf output should look like this:

Next we turn the red hairline "needle" to the direction where the needle is in the calibration image. This is done by altering the zero_angle value (in degrees) so that the needle line points exactly in the needle direction in the calibration image. The value is angle in degrees compared to the picture vertical axis. Straight up would be 0.0 degrees, pointing to the right would be 90.0. Value must be between 0 and 360 degrees. The end result should look like this:

Next we adjust the inner_radius value to form a smaller circle inside the meter circle so that it limits the area where the actual needle arm part moves. The center of the dial will always contain the needle center and is of little use determining the needle angle. To speed up the detection algorithm we limit the centre part out from the detection. After setting the inner circle radius the configuration should look like this. The area where the detection is done is between the green outer circle and the blue inner circle:

Next configure the needle arm geometry in the detection area. This is done by altering two angle values: needle_angle and needle_angle2. The first one tells how wide the needle body is when it hits the outer cirle (measured as an angle when looking from the image center point). A negative angle means the needle does not reach the outer circle, but if lines drawn on the needle egdes would continue, how wide angle would the intersection points on the outer circle make (as a negative value). The second one tells how wide the needle body is in degrees when it hits the inner circle (similarily measured as an angle viewed from the dial center point). When the setting is correct, the red needle edge lines on the calibration image should follow the actual needle edges:

Next, now that the needle geometry is correct, we turn the zero_angle value so that the green dial center line goes exactly through the dial 0-value-point in the meter background. The dial configuration in the example should now look like this and produce the calibration image below.

    @DIAL = 402:253:33:16:-1.0:50.0:3.5:0

Finally for this first dial, determine the direction of rotation with the rot_dir value (0 for clockwise, 1 for counter clockwise). The green arrowhead on the dial outer circle shows the selected direction in the showconf image.

The needle detection algorithm by default uses the average color in the dial center point configured to determine what is the color of the needle being detected from the detection area. However, if the needle centers are of different color as the needle arm or there is some other disturbance in the image (eg. light relection) to prevent determining the color of the needle from the center point, you may need to configure manually the color of the needle. The dial configuration may have additional three optional integer numbers to determine the RGB color value for the needle. In this howto example and with the image setup described here, this is not necessary. Use again your favourite image editing application (GIMP, Paint.NET, etc.) and using the color picker tool on your calibration image find out the RGB-values of the needle part being detected. For example for the dial configured above, we could add the needle color (RGB = 152,57,65) to the configuration and the resulting dial configuration would be like:

    @DIAL = 402:253:33:16:-1.0:50.0:3.5:0:152:57:65

After setting the first and most significant dial, set up the remaining in significance order using the process described above. The dial configuration should look something like this when ready:

Positioning areas

Even the camera would be firmly attached to the meter body, there tends to be small movement between the dials and the camera resulting the detection areas moving slightly in the image. You can help dialEye to deal with this by determining positioning areas in the calibration image that can be used to locate the dials in the meter image.

The positioning areas are listed in the dialEye configuration file using @POSAREA configuration keys. The image may contain from 0 to unlimited amount of positioning areas. However the recommendation is:

• use 3 areas that are clearly visible and spread in different parts of the calibration image
• the positioning areas should not be located just beside the image edge
• use fairly small areas. Large areas are more accurate, but the image processing takes a lot more time when using large ones
• the positioning area boxes should be placed in areas having some recognisable part of the image
• there should not be any changing or moving parts inside the positioning area

The syntax of the @POSAREA-row is:

    @POSAREA=x1:y1:x2:y2

Configure the positioning areas again by trial error or using helpful image tools. showconf visualizes the positioning areas as yellow rectangles on the calibration image.

The sample positioning configuration looks like this and produces the below configuration visualization:

    @POSAREA = 80:130:105:173
    @POSAREA = 448:155:470:183
    @POSAREA = 278:430:296:460

Image "shaking"

If at least one positioning area has been configured, dialEye will by default try to "shake" the metered image (rotate the image and move horisontally and vertically) to find a best match on the positioning areas on the calibration image and the image being detected/metered. The shaking process may be configured to look farther in the image and/or rotate the image more. Larger area and more travel when shaking will consume more time and processor cycles when more positions and angle variations are calculated so depending on your needs you may vary the parameters.

dialEye has been designed to be run periodically and therefore it can save the "shake" result to a file after each detection. Because of the saved state, the shaking parameters can be kept low so that the time shaking consumes is mimimal and the movements are followed by the process all the time because recent shaking results are saved. If however the camera moves a lot quickly (for example it is accidentally bumped into by someone), the shake parameters can be inreased to again find the sync.

The following configuration parameters affect the shaking process:

DISABLE_IMAGE_SHAKE	By default dialEye uses the shake process before reading the dials, if one or more positioning areas are defined. By setting this parameter to true, you can disable the shaking process. (Default: false)
SHAKE_RADIUS	The image may have moved horizontally or vertically. This parameter configures the radius the image is moved to each direction from the center point to find the best position relative to the calibration picture positioning areas. Value 0 disables horisontal and vertical movement. Value 1 is recommended for stable conditions (results to 9 different positions being calculated). Use larger than 1 if you need more processing to locate the dials.
TURN_ANGLE	The metered image is rotated by bottom center point of the image to correct possible camera rotation. This parameter configures the maximum amount in degrees the image is turned to both directions to find best match. Values 1.0 or 0.5 are recommended for stable conditions.
TURN_ANGLE_STEP	This parameters determines the stepping in degrees used in the rotation. Smaller the steps and larger the angle, more angles are calculated and hence longer will the turning take. Value 0.5 is recommended for stable conditions.
SHAKE_FILE	This file is used to save the shaking result for next run. The shake result is saved, if command line parameter -s is given with the meter command. (Default: dialeye_shakes.conf)

Needle detection

The configuration of the dialEye is almost ready. The needle detection algorithm is controlled by two configuration parameters. The parameter DETECTION_TRESHOLD controls the the detection sensitiveness. Again configuring this value requires some trial-error processing. We will try to detect the needles on the clibration image. Detecting/metering is done with command meter and the image begin processed will have to be given as a parameter on the command line:

    python dialEye.py -r meter docs/dialeye_calibration.jpg

The -r option tells dialEye to visualize the detection over the metered image. dialEye generates dialeye_result.png image visualizing the detection. Again adding -g option on graphical environments will open the result image in new window.

The visualization should be checked for two things: are the meter circles located correctly where the dials are and are the needles detected correctly. If the meter circles are not where they should be, the shaking process has not been successful if the camera has moved. After the dials are located correctly concentrate on the red needle detection tracks and the detected green needle angle.

The DETECTION_TRESHOLD value (value between 0-100) can be adjusted so that the needle detection is clear and the green dial angle lines align just where the needles are in the metered image. Below are examples of too small, too large and working treshold.