Comparison of Some Motion Detection Methods in cases of Single and Multiple Moving Objects

Shamir Alavi
International Journal of Image Processing (IJIP), Volume (6) : Issue (5) : 2012 389
Comparison of Some Motion Detection Methods in cases of
Single and Multiple Moving Objects
Shamir Alavi alavi1223@hotmail.com
Electrical Engineering
National Institute of Technology Silchar
Silchar 788010 (Assam), India
Abstract
Motion detection tells us whether there is a change in position of an object with respect to its
surroundings or vice versa. It is applied to various domestic and commercial applications starting
from simple motion detectors to high speed video surveillance systems. In this paper, results
obtained from some simple motion detection algorithms, which use methods like image
subtraction and edge detection, have been compared. The software used for this purpose was
MATLAB 7.6.0 (R2008a). It has been observed that while image subtraction is sufficient to detect
motion in a video stream, combining it with edge detection in different sequences yields different
results in different scenarios.
Keywords: Motion Detection, Edge Detection, Canny, Sobel, Image Subtraction, Computer
Vision
1. INTRODUCTION
Motion detection is a process of confirming a change in position of an object relative to its
surroundings or the change in the surroundings relative to an object [1]. It has paramount
importance in any vision based detection and tracking system. Throughout the last couple of
decades, several techniques have been introduced to accomplish this task effectively. However,
there is no perfect system or method which can overcome the various problems that are faced
during detection. The difficulties are generally associated with lighting condition of the
surrounding, illumination of the object itself which is to be detected, speed of its movement or the
type of object [2].
Generally, motion detection is useful in real time or active video surveillance systems [2]. In this
paper, the main focus is given to the processing of the captured video data to detect motion in it.
Two main methods, image subtraction and edge detection have been used for detection. Three
different cases have been considered in order to compare the results. One detects the motion by
image subtraction only whereas the other two include edge detection in different sequence.
Moreover, two different edge detection techniques, namely Sobel edge detection [3] and Canny
edge detection [4] have also been taken into account while comparing the results. Finally, all of
these are implemented on two different scenarios where the number of moving object is one or
more. After a brief review of image subtraction and the two edge detection techniques, the motion
detection algorithms used here along with comparisons in regard to the scenarios and edge
detection techniques will follow accordingly.
2. Image Subtraction
Image subtraction is one of the popular techniques in image processing and computer vision.
Basically image subtraction can be represented as:
ΔI(i,j) = ICurr(i,j) – IPrev(i,j) (1)
where:

Shamir Alavi
ΔI(i,j) is the difference in image intensity between two consecutive frames. ICurr(i,j) and
IPrev(i,j) represent image intensities for current and previous frames respectively [2].
(a) (b) (c)
FIGURE 1: (a) Frame 1 (b) Frame 2 (c) Image after subtraction
This is primarily done for one of two reasons – leveling uneven sections of an image such as half
an image having a shadow on it, or detecting changes between two images. This detection of
changes can be used to tell if something in the image moved [5].
3. Edge Detection
Edge detection refers to the process of identifying and locating sharp discontinuities in an image.
The discontinuities are abrupt changes in pixel intensity which characterize boundaries of objects
in a scene [6]. The purpose of edge detection in general is to significantly reduce the amount of
data in an image, while preserving the structural properties to be used for further image
processing [7]. Classical methods of edge detection involve convolving the image with an
operator (a 2-D filter), which is constructed to be sensitive to large gradients in the image while
returning values of zero in uniform regions [6]. There are various ways to perform edge detection
as various techniques have been introduced throughout the years. This work will compare two
such techniques while detecting motion which are the Sobel operator or Sobel edge detection [3]
and the Canny edge detection [4]. There are several other operators also such as Prewitt‟s
operator, Robert‟s cross operator, Laplacian of Gaussian etc. but most of them (such as Prewitt‟s
and Roberts‟s operator) work in a fashion similar to Sobel operator [6].
3.1 Sobel Operator
The operator consists of a pair of 3×3 convolution kernels as shown in Figure 1. One kernel is
simply the other rotated by 90°.
-1 0 +1
-2 0 +2
-1 0 +1
Gx Gy
FIGURE 1: Masks used by Sobel Operator
These kernels are designed to respond maximally to edges running vertically and horizontally
relative to the pixel grid, one kernel for each of the two perpendicular orientations. The kernels
can be applied separately to the input image, to produce separate measurements of the gradient
component in each orientation (let us call these Gx and Gy). These can then be combined to find
the absolute magnitude of the gradient at each point and the orientation of that gradient [3]. The
gradient magnitude is given by:
(2)
Typically, an approximate magnitude is computed using:
+1 +2 +1
0 0 0
-1 -2 -1

Shamir Alavi
(3)
which is much faster to compute.
The angle of orientation of the edge (relative to the pixel grid) giving rise to the spatial gradient is
given by:
(4)
3.2 Canny Edge Detection
The algorithm runs in five separate steps [7]:
i. Smoothing: Blurring the image to remove noise.
ii. Finding gradients: The edges are marked where the gradients of the image has
large magnitudes.
iii. Non-maximum suppression: Only local maxima are marked as edges.
iv. Double thresholding: Potential edges are determined by thresholding strong
and weak edges.
v. Edge tracking by hysteresis: Final edges are determined by suppressing all the
edges that are not connected to a very certain (strong) edge.
(a) (b) (c)
FIGURE 2: (a) Original image (b) Sobel edge (c) Canny edge
4. Algorithms Used for Motion Detection
Using image subtraction and edge detection as the main tool, following three algorithms have
been used:
4.1 Image Subtraction Method
The first task is to extract frames from the continuous video stream so that they can be processed
further for our next tasks. The steps for this algorithm are stated below:
i. Extract frames from video stream.
ii. Write the extracted frames as image files.
iii. Subtract the previous image from current image as stated in (1).
iv. Convert image to binary image.
v. Label connected components.
vi. Perform blob analysis (i.e. measure properties of each labeled image regions).
vii. Calculate the centre of mass of each labeled region and label it (to detect as
many moving elements as possible).
viii. Play the labeled images as a continuous video stream to detect motion.

Shamir Alavi
4.2 Edge Detection after Image Subtraction
In this algorithm, edge detection is performed after image subtraction. The algorithm is as follows:
iii. Subtract the previous image from current image.
iv. Convert the image to grayscale.
v. Detect edges.
vi. Label connected components.
vii. Perform blob analysis.
viii. Calculate the centre of mass of each labeled region and label it.
ix. Play the labeled images as a continuous video stream to detect motion.
4.3 Image Subtraction after Edge Detection
In this case, edge detection is performed before image subtraction. The algorithm is stated below:
iii. Detect edges in all the images.
iv. Subtract the previous image from current image.
v. Convert subtracted image to binary image.
vi. Label connected components.
vii. Perform blob analysis.
viii. Calculate the centre of mass of each labeled region and label it.
ix. Play the labeled images as a continuous video stream to detect motion.
5. Comparison of the Motion Detection Algorithms
5.1 Considered Scenarios
Two different scenarios have been considered for comparison:
Scenario 1: A man is walking while everything else is still. Therefore, we have a single moving
object in this scenario (as in Figure 1)
Scenario 2: Plastic caps are being collected from one conveyor belt to another. Here, the caps
as well as the belts are in motion. Hence, we have multiple moving objects in this scenario (as in
Figure 2).
5.2 Visual Comparison
From the extracted frames, two consecutive frames have been taken for comparison and
analysis:
(a) (b) (c) (d) (e)
(f) (g) (h) (i) (j)

Shamir Alavi
(k) (l) (m)
FIGURE 3: Comparison of motion detection algorithms on scenario 1, (a) and (b) two consecutive frames (c)
subtracted image (d) algo
*
4.1 (e) algo 4.2 – Sobel (thresh = 0.17) (f) algo 4.2 – Canny (thresh = 0.17) (g)
algo 4.2 – Canny (thresh = 0.55) (h) subtracted image of Sobel edges (i) algo 4.3 – Sobel (thresh = 0.17) (j)
subtracted image of Canny edges (thresh = 0.17) (k) algo 4.3 – Canny (thresh = 0.17) (l) subtracted image
of Canny edges (thresh = 0.55) (m) algo 4.3 – Canny (thresh = 0.55)
*
algo = algorithm
(a) (b) (c) (d) (e)
(f) (g) (h) (i) (j)
(k) (l) (m)
FIGURE 4: Comparison of motion detection algorithms on scenario 2, (a) and (b) two consecutive frames (c)
subtracted image (d) algo 4.1 (e) algo 4.2 – Sobel (thresh = 0.17) (f) algo 4.2 – Canny (thresh = 0.17) (g)
algo 4.2 – Canny (thresh = 0.55) (h) subtracted image of Sobel edges (i) algo 4.3 – Sobel (thresh = 0.17) (j)
subtracted image of Canny edges (thresh = 0.17) (k) algo 4.3 – Canny (thresh = 0.17) (l) subtracted image
of Canny edges (thresh = 0.55) (m) algo 4.3 – Canny (thresh = 0.55)
N.B.: For Canny edges, higher threshold = thresh, lower threshold = 0.4 * thresh, standard
deviation of the Gaussian filter = 1. For instance, in figure 4(j), higher threshold = 0.17, lower
threshold = 0.4 * 0.17 = 0.068, standard deviation of the Gaussian filter = 1.
5.3 Comparison Tables
The following table summarizes the visual comparisons shown above:
Algorithm Edge Scenario 1 Scenario 2
4.1
Detects motion quite well despite
data loss due to binary conversion
of the subtracted image - fig 3(d)
Partially detects motion of only one
cap - fig 4(d).

Shamir Alavi
TABLE 1: Comparison of the motion detection algorithms under different scenarios
Average
**
time taken to obtain the above shown figures (only the ones in which motions are
detected):
TABLE 2: Comparison of average time taken by the algorithms to detect motion
**
Each program has been run 5 times on the same system (Intel®
Core(TM) i5 CPU M430 @ 2.27
GHz, 4GB DDR3 RAM, ATI Radeon HD 5400, Win 7 64-bit) and their average was considered.
***
Total number of frames in the video stream for this scenario = 80
****
Total number of frames in the video stream for this scenario = 249
6. Result and Discussion
From the comparisons shown above in table 1, it is clear that image subtraction only is good
enough to detect motion in case of a single moving object in front of a still background. However,
it fails to accomplish this task properly when there are multiple moving objects.
4.2
Sobel Minor visual detection – fig 3(e). Minor visual detection – fig 4(e).
Canny
(thresh=0.17)
Great visual detection (detects
movement of the whole body) – fig
3(f).
Unable to detect the motion of the
belts (but better than Sobel as in fig
4(e)) – fig 4(f).
Canny
(thresh=0.55)
Detection of strong edges only
(better than Sobel as in fig 3(e)) –
fig 3(g).
Unable to detect the upper cap and
the belts (still better than Sobel as in
fig 4(e)) – fig 4(g).
4.3
Sobel
Detection quality almost similar to
that of algo 4.1 -fig 3(i).
Partially detects both the caps but
cannot detect movements of the belts
– fig 4(i).
Canny
(thresh=0.17)
Detects movement of the man but
erroneously detects several
portions of the still pavement as
moving objects – fig 3(k).
Great visual detection – detects
movement of both the caps and the
lower belt; slightly detects movement
of the upper belt as well – fig 4(k).
Canny
(thresh=0.55)
Detection of strong edges only (no
false detection) – fig 3(m).
However, detection quality is lower
than that of Sobel detection as in
fig 3(i).
Unable to detect the motion of the
belts – fig 4(m). However, detection
quality is better than that of Sobel
detection as in fig 4(i).
Algorithm Edge
Scenario1
***
Scenario 2
****
Figure Average time(sec) Figure Average time(sec)
4.1 3 (d) 3.0717644 4 (d) 4.1041660
4.2
Sobel 3 (e) 3.7338478 4 (e) 6.4473798
Canny
(thresh
= 0.17)
3 (f) 4.0066358 4 (f) 6.4270532
Canny
(thresh
= 0.55)
3 (g) 3.8619630 4 (g) 6.3946814
4.3
Sobel 3 (i) 7.5335376 4 (i) 13.8421282
Canny
(thresh
= 0.17)
3 (k) 13.3262350 4 (k) 26.4577750
Canny
(thresh
= 0.55)
3 (m) 12.6463506 4 (m) 25.0108116

Shamir Alavi
For a single moving object, the best result has been obtained by performing Canny edge
detection after image subtraction where a low threshold value was used. Sobel edge detection
could not perform as well as Canny edge detection. On the other hand, in case of multiple moving
objects, the best result came from performing Canny edge detection before image subtraction.
Here also, a low threshold value was used.
From table 2, we can see that „Image Subtraction Method‟ is the fastest of the three motion
detection methods that are compared here. Between the other two methods, performing edge
detection after image subtraction is substantially faster. A careful look at the average times taken
by this method also tells us that it is not that slow in comparison to the simpler image subtraction
method (extra time taken by algo 4.2 as compared to 4.1 is below 1 second for scenario 1 and
just above 2 seconds for scenario 2). For multiple moving objects, though Canny gave the best
result, it came at the cost of high computation time (26.4577750 seconds).
7. Conclusion and Future Research
To sum up, edge detection, in addition to image subtraction is necessary to detect motion
properly. Canny edge detection, in spite of being computationally slower and more expensive as
a result, gives the best outcome under any circumstances. On top of it, „Image Subtraction after
Edge Detection‟ is the best method out of the three discussed above if we can compromise the
higher computational time taken by it.
For further research, only Canny edge detection will be used as it has been deemed to work best
among all other edge detection methods available at present [8, 9] and is referred to as a
‘modern standard’ [10]. It is also necessary to improve the higher computation times with Canny
edge detection, especially in complex situations with multiple moving objects. This research work
can be further extended by testing under different lighting conditions, differing the distance of the
moving objects and finally going live with a faster algorithm and system.
8. References
[1] Wikipedia. "Motion Detection." Internet: http://en.wikipedia.org/wiki/Motion_detection,
May 20, 2012 [Jun. 16, 2012].
[2] R. B. Wahyu, Tati R. Mengko, Bambang Pharmasetiawan and Andryan B. Suksmono.
"Motion Detection Using Image Subtraction Edges Detection." Risalah Lokakarya
Komputasi dalam Sains dan Teknologi Nuklir XVII, pp. 157-171, Aug. 2006.
[3] J. Matthews. “An Introduction to Edge Detection: The Sobel Edge Detector.” Internet:
http://www.generation5.org/content/2002/im01.asp, 2002.
[4] J. Canny. “A Computational Approach to Edge Detection.” IEEE Transaction on Pattern
Analysis and Machine Intelligence, vol. PAMI-8, no. 6, pp. 679-697, Nov. 1986
[5] Wikipedia. "Image subtraction." Internet: http://en.wikipedia.org/wiki/Image_subtraction,
Sep. 27, 2011 [Jun. 16, 2012].
[6] R. Maini and Dr. H. Aggarwal. "Study and Comparison of Various Image Edge Detection
Techniques." International Journal of Image Processing (IJIP), vol. 3, iss. 1, pp. 1-11,
Feb. 28, 2009.
[7] "Canny Edge Detection." Internet:
http://www.cvmt.dk/education/teaching/f09/VGIS8/AIP/canny_09gr820.pdf, Mar. 23,
2009.

Shamir Alavi
[8] E. Nadernejad, S. Sharifzadeh and H. Hassanpour. "Edge Detection Techniques:
Evaluations and Comparisons." Applied Mathematical Sciences, vol. 2, no. 31, pp. 1507 –
1520, 2008.
[9] M. Juneja and P. S. Sandhu. "Performance Evaluation of Edge Detection Techniques for
Images in Spatial Domain." International Journal of Computer Theory and Engineering,
vol. 1, no. 5, pp. 614-621, Dec. 2009.
[10] M. Heath, S. Sarkar, T. Sanocki and K. Bowyer. "Comparison of Edge Detectors: A
Methodology and Initial Study." Computer Vision And Image Understanding, vol. 69, no.
1, pp. 38–54, Jan. 1998.

Comparison of Some Motion Detection Methods in cases of Single and Multiple Moving Objects

More Related Content

What's hot (20)

Similar to Comparison of Some Motion Detection Methods in cases of Single and Multiple Moving Objects (20)

Recently uploaded (20)

Comparison of Some Motion Detection Methods in cases of Single and Multiple Moving Objects