Analysis of Error in Image Logging between Subsequent Frames of a Streaming Video

Mr. THANMAY J.S

Assistant Professor, Mechanical Department, G.S.S.S.I.E.T.W Mysore.

Email: thanmay26@rediffmail.com

Abstract— In Image Analysis of a streaming video for certain

applications, it may not be necessary to log every frame provided

by an image acquisition device. In fact, it may be more practical

and resourceful to log frames at certain intervals. To log frames

at a certain interval acquisition device must be configured for

Frame Grab Interval, Trigger property, Frame Rate property,

umber of frames to be logged etc. Configuring the property to

an integer value specifies which frame should be logged.

Generally error inclusions due to delay in logging data is a

critical issue which is well understood in this report. Set of

experimental results with changes in Frame rate and umber of

Frames to be logged along with percentage errors are analyzed in

this report

Whenever captured image is consider for visual analysis

traveling on a fast-moving conveyor belt. The image capture

system must know when they are in the right place and capturing

the image and transmit the result for processing before the next

sample arrives. Selection of trigger sensors is no problem; the

next consideration is timing which can be a sporadic in nature

and depends upon logging data and error in acquisition. This

report explains the importance of adjusting the device parameters

like frame rate and grab interval at which frames were logged

and how number of frames logged per second with error in

logging frames.

Keywords: Image Acquisition, Video Device setting, Accessing

Device Properties, Frame Rate, Grab Interval, Logging Data,

Time difference between frames etc.

I. I�TRODUCTIO� TO IMAGE ACQUISITIO�

This document provides an information for the Acquiring

objects that represent the connection to the device. This kind

of Image Acquisition system is well suitable for Artificial

intelligence and Machine vision system in industries. image

acquisition application involves these major steps:

Step 1) Starting the video input object -- You start an object by

calling the start function. Starting an object prepares the object for

data acquisition. For example, starting an object locks the values

of certain object properties (they become read only). Starting

an object does not initiate the acquiring of image frames,

however. The initiation of data logging depends on the execution

of a trigger. The example calls the start function to start the

video input object. Objects stop when they have acquired the

requested number of frames.

Step 2) Triggering the acquisition -- To acquire data, a video

input object must execute a trigger. Triggers can occur in

several ways, depending on how the Trigger Type property is

configured. For example, if you specify an immediate trigger,

the object executes a trigger automatically, immediately after it

starts. If you specify a manual trigger, the object waits for a

call to the trigger function before it initiates data acquisition.

Step 3) Bringing data into the workspace -- The Image analysis

software stores acquired data in a memory buffer, a disk file,

or both, depending on the value of the video input object

Logging Mode property. To work with this data, you must

bring it into the workspace. Once the data is in the workspace,

you can manipulate it as you would any other data.

II. BASIC IMAGE ACQUISITIO� PROCEDURE

of Image Data after you create the video input object and configure its properties. The Image Acquisition system enables

you to connect to an image acquisition device from within a

software interface session and based on object technology, the

Image Acquisition system provides functions for creating

Mr. THA�MAY J.S, Assistant Professor in Mechanical Department, G.S.S.S

Institute of Engineering and Technology for Women, K.R.S. Road,

Metagalli, Mysore-570016, Karnataka, India.

(E-mail of corresponding author: thanmay26@rediffmail.com).

This section illustrates the basic steps required to create an

image acquisition application by implementing a simple

motion detection application. The application detects

movement in a scene by performing a pixel-to-pixel

comparison in pairs of incoming image frames. If nothing

moves in the scene, pixel values remain the same in each

frame. When something moves in the image, the application

displays the pixels that have changed values.

To use the Image Acquisition system to acquire image data,

you must perform the following basic steps:

Step 1: Install and configure your image acquisition device

Step 2: Configure image acquisition properties

Step 3: Create a video input object Step 4: Preview the video stream Step 5: Acquire image data

Step 6: doing Image analysis and displaying results

Step 7: Cleaning up the memory

Certain steps are optional and this kind of system to acquire

streaming images is well suitable for Machine vision system.

III. IMAGE ACQUISITIO� PROPERTIES

1) The frame rate describes how fast an image acquisition

device provides data, typically measured as frames per second

(Figure �o 1). Devices that support industry-standard video

formats must provide frames at the rate specified by the

standard. For RS170 and �TSC, the standard dictates a frame

rate of 30 frames per second (30 Hz). The CCIR and PAL

standards define a frame rate of 25 Hz. �on-standard devices

can be configured to operate at higher rates. Generic Windows

image acquisition devices, such as Webcams, might support

many different frame rates. Depending on the device being

used, the frame rate might be configurable using a device-

specific property of the image acquisition object. The rate at

which the Image Acquisition software can process images

depends on the processor speed, the complexity of the

processing algorithm, and the frame rate. Given a fast

processor, a simple algorithm, and a frame rate tuned to the

acquisition setup, the Image Acquisition software can process

data as it comes in.

2) The Frame Grab Interval property specifies how often

the video input object acquires a frame from the video stream

(Figure �o 1). By default, objects acquire every frame in the

video stream, but you can use this property to specify other

acquisition intervals. For example, when you specify a Frame

Grab Interval value of 3, the object acquires every third frame

from the video stream, as illustrated in this figure. The object

acquires the first frame in the video stream before applying the

Frame Grab Interval.

3) There are two methods to use Trigger Properties one is

Automatic and other is Manual (Figure �o 1). To use an

Automatic immediate trigger, simply create a video input

object. Immediate triggering is the default trigger type for all

video input objects. With an immediate trigger, the object

executes the trigger immediately after you start the object

running with the start command. To use a manual trigger,

create a video input object and set the value of the Trigger

Type property to 'manual'. A video input object executes a

manual trigger after you issue the trigger function.

4) Frames per Trigger are used to move multiple frames of

data from the memory buffer into the workspace (Figure �o 1).

By default, getting data retrieves the number of frames

specified in the Frames per Trigger property but you can

specify any number. In the Image Acquisition system, one can

specify the amount of data intended to be acquired as the

number of frames per trigger. You specify the desired size of

your acquisition as the value of the video input object Frames

per Trigger property.

Figure �o 1. Image Acquisition properties in a video stream.

Figure �o 2. Machine Vision application using video stream.

The question arises is “why to understand all these things?” the

reason is “Timing”. Observe the above figure (Figure �o 2)

where moving items are constantly monitored by Machine

vision. If logging time and process does not match each other

then caps of a bottle will be places on label and label will be

stuck on caps. Is this possible? “YES”; for this I give you a

following popular experiment.

IV. U�DERSTA�DI�G PROPERTIES OF IMAGE ACQUISITIO�

Experiment �umber: 01

The following experiment is done by lighting a match stick in

front of a web cam with 30 frames per second (Figure �o 3). This

gives an idea how well the image is acquired. 30 frames per

second means 1 frame is 0.033 second. If I calculate

number of frames then I should get exact timing of a match

stick burning rate. Let us see how?

Figure �o 3. Video stream of a Match stick burnt with 30 fps and

number of frames logged is 400.

Please observe the above figure (Figure �o 3); 30 fps means

0.033 seconds per frame then the burning of match stick is 146

frames. This gives us the Idea 146/30 is the time required for a

match stick to burn that means 4.866 seconds?

Experiment �umber: 02

The following experiment is done by lighting a match stick in

front of a web cam with 30 frames per second (Figure �o 4) and

number of frames logged is 300. it appears 155 frames shows

burning rate then 155/30 should be actual burning rate of match

stick which is 5.166 seconds?.

Figure �o 4. Video stream of a Match stick burnt with 30 fps and

number of frames logged is 300.

�ote: Every match stick can burn up to 25 seconds.

V. REASO�S FOR PROPERTIES OF IMAGE ACQUISITIO� TO

DISOBEY

When ever there is a queue of data trying to enter the

processor of a system they hung up due to low in processing

speed. This gives rise to an error which is known as “logging

error between two frames”.

One frames which waits to be logged on delays another

frame to wait until it is processed. Once it is processed the

waiting frame does not counts as the video stream is a dynamic

system where data I logged on the basis of number of frame to be

logged. For a perfect flow of data to log and to process the

system understands few rules they are as follows.

Following set of experiments is as shown with their

corresponding results in Tabular forms. Variations to all needs

are specifically noted and the results are tabulated.

VI. EXPIRIME�T CASE STUDY FOR ERROR A�ALYSIS BETWEE�

LOGGI�G OF TWO FRAMES

The following three tables shows the set of results due to

variation in either frame rate per second or variation in number of

frames to be logged or constants in them.

TABLE I

EXPIRIME�T �O: 01

To study variations in frame rate per second and corresponding variations in

number of frames to be logged.

Experiment Frame �umber Logging Average Percent

Rule �umber: 01 �o Rate of rate Diff Error

Logging Rate = 1/logged Per Sec Log per frames

i.e. 30 fps = 1/30 log per sec i.e. 0.033 sec per frame second logged

Rule �umber: 02

1 30 300

0.03 0.1474 11.4098

2 25 250 0.04 0.1473 10.7297

Average difference between each frame = first (n-1) frame ~ (n) second frame

3 20 200 0.05 0.1458 9.5844

Rule �umber: 03 Percentage Error = logging Rate-Average

Difference * 100

If these errors are not noted then the results are those which

you have just seen. �ow let us solve those two experiments.

4 15 150 0.06 0.1463 7.9682 5 10 100 0.1 0.1422 4.2212 6 5 50 0.2 0.1341 6.5878

TABLE II

EXPIRIME�T �O: 02

To study constant frame rate per second and variation in number of frames to

be logged.

[1] Experiment umber: 01

�umber of frames per second = 30 fps

�umbers of frames to be logged = 400 f

Time based acquired frames = 146 f

Experiment Frame �umber

�o Rate of

Log frames

per logged

second

1 30 300

Logging Average Percent Error

rate Diff

0.033 0.1474 11.4098

Average difference between two frames= 0.1456 sec 2 30 250 0.033 0.1472 11.3888

Error between two frames (Average) = 11.2231

Time for match stick burnt (Approximate) = 20sec sec Then the exact time should be 146*0.112231+4.86= 21.245726 Sec

[2] Experiment umber: 02

�umber of frames per second = 30 fps

�umbers of frames to be logged = 300 f

Time based acquired frames = 155 f

Average difference between two frames= 0.1464 sec

Error between two frames (Average) = 11.3078

Time for match stick burnt (Approximate) = 23sec sec Then the exact time should be 155*0.113078+5.16= 22.68709 Sec

�ow seeing this result which shows up to nano decimal is

astounding. These results can bring forth the Machine Vision

system to its timing but the conclusion cannot be easily done.

What if number of frames to be logged is 5 to 5000 or what if

number of frames per second is 5 to 30? These kinds of

questions need specific answers. For this several experiments

are done to know the suitable answer. The variation in frames

per second and variation in frames to be logged are constantly

varied and errors in logging are studied. This showed many

important factors which cannot be explained in singular terms.

3 30 200 0.033 0.1469 11.3541

4 30 150 0.033 0.1454 11.2056

5 30 100 0.033 0.1438 11.0495

6 30 50 0.033 0.1348 10.1422

TABLE III

EXPIRIME�T �O: 03

To study variations in frame rate per second and constant number of frames to

be logged.

Experiment Frame �umber Logging Average Percent Error

�o Rate of rate Diff Log frames

per logged

second

1 30 100 0.03 0.1428 10.9444

2 25 100 0.04 0.1416 10.1646

3 20 100 0.05 0.1412 9.1232

4 15 100 0.06 0.1414 7.4768

5 10 100 0.1 0.1357 3.5737

6 5 100 0.2 0.1413 5.8747

Observe in each table the error percentage shown. What

variation can anyone observe in them unless they are shown as

in �umerical values? Remember timing is a question to be

answered. In each of the above three table logging rate is not

our problem this is because intense image has larger logging

rate as every thigh to be logged is in the rate of pixels. Then

the error registered is our main picture. Averaging the logging

rate revokes all our problems and to obtain perfect timing error

percentage is calculated. Machine Vision, Computer Vision or

Image Analysis of any Video stream requires logging of data if the

data has error then the whole Analysis goes wrong.

VII. CO�CLUSIO�

How can any one believe a match stick can burn only for 3 or

5 second? This shows even video stream registration need

processing speed and this has to be concluded in standards.

The above Tabulation results are now shown of graphical

forms to conclude. Observe the below three bar chart which

clarify all our doubt regarding image acquisition and its error

incurred.

EXPIRIME�T �O: 01

To study variations in frame rate per second and corresponding variations in

number of frames to be logged.

350

300

250

200 Frames per second

Percentage Error

150 Number of frames logged

100

50

0

1 2 3 4 5 6

EXPIRIME�T �O: 02

To study constant frame rate per second and variation in number of frames to

be logged.

350

300

250

200 Frames per second

Percentage Error

150 Number of frames logged

100

50

0

1 2 3 4 5 6

EXPIRIME�T �O: 03

To study variations in frame rate per second and constant number of frames to

be logged.

120

100

80

Frame Per Second

60 Percentage Error

Number of frames logged

40

20

0

1 2 3 4 5 6

EXPIRIME�T �O: 01 CO�CLUSIO�

Higher frame rate per second and larger frames to be logged

always shows higher error. Decreasing both simultaneously

reduces error progressively.

EXPIRIME�T �O: 02 CO�CLUSIO�

Higher frame rate per second and higher frames to be logged

always shows higher error. Decreasing only frame to be logged

and making frames per second constant does not reduce error

much.

EXPIRIME�T �O: 03 CO�CLUSIO�

Variation in frame rate per second and frames to be logged

remains constant shows error reduction dramatically.

Final conclusion makes one and all to remember, that

acquiring image data from an video stream does not gives

accurate timing unless error between logging of two

subsequent frames are described. Increasing or decreasing

frame rate per second or number of frames to be logged cannot

be ignored.

Machine Vision attached with Robotics should be

able to detect and understand the multitude of tasks that can

arise in “hard engineering” in industries, where low-tolerance

features giving rise to highly variable images to be processed

along with timing of work handling. What if the timing goes

wrong then the learning ability of artificial intelligence will

start to learn the trial and error method which humans have

mastered it. The culture of developing Machine Vision or the

Artificial Intelligence should not be blinded with error in

Image acquisition and error in processing logged data. If

logging data depends on processor speed then perfection in

analysis of a Image from a streams of data has to answer

average error in processing them, this will lead to a perfection.

VIII. REFERE�CES

References are gathered through primary data of literature

survey through internet sites like http://citeseerx.ist.psu.edu

and by interacting with experts. References and readily

available publications are used only for knowledge purpose

but the experiments and proof for the report are self build.

Samples of the collective formats for various types of

references are given below. [1] A. Said, W. A. Pearlman, “A �ew Fast and Efficient Image Codec

Based on Set Partitioning in Hierarchical Trees”, IEEE Trans. Circuits

and Syst. Video Technol., 6(3), 243-250 (1996).

[2] P. Corriveau, A. Webster, "VQEG Evaluation of Objective Methods of

Video Quality Assessment", SMPTE Journal, 108, 645-648, 1999.

[3] A.M. Rohaly, P. Corriveau, J. Libert, A. Webster, V. Baroncini, J.

Beerends, J.L Blin, L. Contin, T. Hamada, D. Harrison, A. Hekstra, J.

Lubin, Y. �ishida, R. �ishihara, J. Pearson, A. F. Pessoa, �. Pickford,

A. Schertz, M. Visca, A. B. Watson, S. Winkler: "Video Quality Experts

Group: Current results and future directions." Proc. SPIE Visual

Communications and Image Processing, vol. 4067, Perth, Australia,

June 21-23, 2000.

[4] C. B. Lambrecht, Ed., “Special Issue on Image and Video Quality

Metrics”, Signal Processing, vol. 70, (1998).

[5] A. Said and W. A. Pearlman, “A new, fast, and efficient image codec

based on set partitioning in hierarchical trees,” IEEE Trans. Circuits

Syst. Video Technol., vol. 6, pp. 243-250, June 1993.

[6] “Jpeg software codec.” Portable Research Video Group, Stanford

University, 1997. Available via anonymous ftp from

havefun.stanford.edu:pub/jpeg/JPEGv1.1.tar.Z.

[7] J. Ashley, R. Barber, M. Flickner, J. Hafner, D. Lee, W. �iblack, and

D. Petkovic. Automatic and semi-automatic methods for image

annotation and retrieval in QBIC. In W. �iblack and R. C. Jain, editors,

Storage and Retrieval for Image and Video Databases III. SPIE - The

International Society for Optical Engineering, 1995

[8] A. Rav-Acha and S. Peleg. Restoration of multiple images with motion

blur in different directions. IEEE Workshop on Applications of

Computer Vision, 2000. [9] K. Cinkler and A. Mertins, \Coding of digital video with the edge-

sensitive discrete wavelet transform," in IEEE International

Conference on Image Processing, vol. 1, pp. 961-964, 1996.

[10] J. Bach, C. Fuler, A. Gupta, A. Hampapur, B. Horowitz, R. Humphrey,

R. Jain, and C. Shu, “The Virage Image Search Engine: An Open

Framework for Image Management,” Proc. SPIE Conf. Storage and

Retrieval for Image and Video Databases IV, vol. 2670,

[11] pp. 76-87, 1996. R. J. Vidmar. (1992, Aug.). On the use of atmospheric

plasmas as electromagnetic reflectors. IEEE Trans. Plasma Sci.

[Online]. 21(3), pp. 876-880. Available:

http://www.halcyon.com/pub/journals/21ps03-vidmar