YOU ARE DOWNLOADING DOCUMENT

Please tick the box to continue:

Transcript
Page 1: 0-0. The Detection of Persons in Cluttered Beach Scenes Using Digital Video Imagery And Neural Network-Based Classification.pdf

March 27, 2007 14:53 WSPC/157-IJCIA 00192

International Journal of Computational Intelligence and ApplicationsVol. 6, No. 2 (2006) 149–160c© Imperial College Press

THE DETECTION OF PERSONS IN CLUTTERED BEACHSCENES USING DIGITAL VIDEO IMAGERY AND NEURAL

NETWORK-BASED CLASSIFICATION

STEVE GREEN and MICHAEL BLUMENSTEIN

School of Information and Communication TechnologyGriffith University, Gold Coast

Queensland 9726, Australia

MATTHEW BROWNE

CSIRO Mathematical and Information SciencesCleveland, Queensland 4163, Australia

RODGER TOMLINSON

Griffith Centre for Coastal Management, Griffith UniversityGold Coast, Queensland 9726, Australia

Received 31 January 2006Revised 8 June 2006

This paper presents an investigation into the detection and quantification of personsin real-world beach scenes for the automated monitoring of public recreation areas.Aside from the obvious use of video and digital imagery for surveillance applications,this research focuses on the analysis of images for the purpose of predicting trends inthe intensity of public usage at beach sites in Australia. The proposed system usesimage enhancement and segmentation techniques to detect objects in cluttered scenes.Following these steps, a newly proposed feature extraction technique is used to representsalient information in the extracted objects for training of a neural network. The neuralclassifier is used to distinguish the extracted objects between “person” and “non-person”categories to facilitate analysis of tourist activity. Encouraging results are presented forperson classification on a database of real-word beach scene images.

Keywords: People detection; image segmentation; modified direction feature; video imageanalysis; beach imagery.

1. Introduction

This paper describes a novel person detection system for analyzing beach sceneimagery. Quantifying people on beaches can provide valuable information for localauthorities to estimate the number of persons using a beach on a particular day.People use beaches for exercise, relaxation, and social activities. The monitoring oflocal beach behavior can provide valuable information regarding whether currentamenities at a particular site are sufficient to meet changing levels of demand. Mostmajor beaches around Australia, and the world, have World Wide Web cameras

149

Int.

J. C

omp.

Int

el. A

ppl.

2006

.06:

149-

160.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

DA

DE

DE

SA

O P

AU

LO

on

05/2

1/13

. For

per

sona

l use

onl

y.

Page 2: 0-0. The Detection of Persons in Cluttered Beach Scenes Using Digital Video Imagery And Neural Network-Based Classification.pdf

March 27, 2007 14:53 WSPC/157-IJCIA 00192

150 S. Green et al.

(sometimes also called web cams) that provide local beach information accessiblethrough the Internet. Coastalwatcha Pty. Ltd. provides this service in Australia viaa national network of remote cameras.

The remote video feeds that Coastalwatch provide offer up-to-date informationon surf and current weather conditions at many beach locations around Australia.The streamed images from Coastalwatch web cams are of a low resolution. There-fore, aside from the inherent difficulties in dealing with variable outdoor imagery,processing low quality images provides quite a challenge for any automated humanquantification system. The low quality beach imagery is one of the main constraintsupon this research. People (and other objects) within these images can, in somecases, only be 10 or 15 pixels in height.

Another important aspect of detecting people on beaches is that of safety.Beaches can provide a dangerous environment for people of all ages.1 Currently, life-guards provide assistance for swimmers who encounter trouble between the flags,which are designated as patrolled areas. This requires human surveillance of thebeach to notify a lifeguard when a swimmer is distressed or in trouble. If a sys-tem could be developed to monitor and detect erratic or uncharacteristic swimmerbehavior and notify the lifeguard, this would provide extra safety measures forswimmers.

The remainder of this paper is organized into four main sections. Section 2 givesan overview of existing techniques in the literature relating to object and persondetection, Sec. 3 provides a detailed description of the proposed person detectionsystem, in Sec. 4, the results attained using the proposed system are presented andfinally conclusions and future work are provided in Sec. 5.

2. Overview of Existing Techniques for Person and ObjectDetection

The automated detection of persons and their behavior in beach scenes is a novelapplication in the field of video surveillance. However, a number of techniques andsystems have been proposed for automated analysis of humans in other indoor andoutdoor situations. Some of these are reviewed and detailed in the paragraphs thatfollow.

Schofield et al.2 proposed a system for analyzing video imagery to count personson the floors of buildings for improving the efficiency of elevator systems. They haveused intelligent techniques for identifying the background of a scene.3 The limitationof this system was that it was based indoors. However, a number of systems havebeen proposed for dealing with outdoor conditions, which have proved to be farmore variable. Iketani et al.4 propose a system for real-time detection of intruders(persons) in difficult, outdoor and cluttered scenes using information from videoimagery over space and time. Sacchi et al.5 present advanced image processing tools

ahttp://www.coastalwatch.com

Int.

J. C

omp.

Int

el. A

ppl.

2006

.06:

149-

160.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

DA

DE

DE

SA

O P

AU

LO

on

05/2

1/13

. For

per

sona

l use

onl

y.

Page 3: 0-0. The Detection of Persons in Cluttered Beach Scenes Using Digital Video Imagery And Neural Network-Based Classification.pdf

March 27, 2007 14:53 WSPC/157-IJCIA 00192

Detection of Persons in Cluttered Beach Scenes 151

for remote monitoring of a tourist site involving the counting of persons in outdoorscenes. Bartolini et al.6 propose a system for automatically counting persons gettingin and off a bus using image sequence analysis for allocating appropriate resourceson bus lines. Pai et al.7 present a system for pedestrian tracking using vision-basedtechniques to prevent traffic accidents. Finally, person tracking in complex sceneshas emerged as a challenging problem in the area of video surveillance. A numberof systems have recently been proposed in the literature, which address this field ofresearch.8–10

Aside from the quantification of persons, some studies have also been performedfor the purpose of quantifying and tracking the behavior of motor vehicles. Thesestudies (mainly dealing with outdoor imagery) are of relevance to the present studyas the techniques that are detailed for detection, monitoring and classification aretransferable to tracking human objects in outdoor scenes.

Tai et al.11 present an image tracking system, which locates motorcycles andother vehicles for traffic monitoring and accident detection at road intersections.Another system recently proposed by Ha et al.12 has employed a neural network-based edge detector for vehicle detection and traffic parameter estimation (vehiclecount, class and speed) in an image-based traffic monitoring system. Other systemssuch as that of Wohler and Anlauf,13 have been used to assist drivers in automobilesthrough the detection of overtaking vehicles. They employed an adaptable time-delay artificial neural network (ANN) to analyze complete image sequences. Finally,some researchers have utilized additional indicatory information for the purposeof vehicle detection. For example, Altmann et al.14 have developed a system todetect military vehicles using acoustic and seismic information for application indisarmament and peacekeeping.

A number of current object detection systems, operating within static imagery,employ an exhaustive search technique to locate regions of interest.15–17 Theexhaustive search technique is computationally expensive, and therefore eliminat-ing regions of non-interest is important to reduce not only the false positive rate,but also to reduce the computation time of the search. Our proposed system cur-rently reduces this search time by removing background information, and then onlysearches foreground regions that seem promising. Further details pertaining to theproposed system are discussed in the next section.

3. System Overview

This research describes a model for the automatic detection of people on beaches.The proposed system uses a classification-based strategy, which searches gray-scale images of beach scenes for potential objects of interest. The sub-imagesextracted are then processed to determine whether a person object has been found.Each sub-image is processed by using an edge detector, feature extractors andfinally a neural-based classifier. An overview of the entire system is presented inFig. 1.

Int.

J. C

omp.

Int

el. A

ppl.

2006

.06:

149-

160.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

DA

DE

DE

SA

O P

AU

LO

on

05/2

1/13

. For

per

sona

l use

onl

y.

Page 4: 0-0. The Detection of Persons in Cluttered Beach Scenes Using Digital Video Imagery And Neural Network-Based Classification.pdf

March 27, 2007 14:53 WSPC/157-IJCIA 00192

152 S. Green et al.

Fig. 1. System overview.

3.1. Object detection and segmentation

The first important stage for segmenting objects out of a complex scene is to detectthe object boundaries in the image. Boundaries encompass an object and are regionswhere there is a high change in luminance values. Beach scenes mainly contain largeareas of sand and sea. From one point of view, this is of great benefit when detectingpeople, as people in beach scenes contrast quite significantly with the sand andtherefore are easier to detect, than say, in a forested scene. However, conversely,the presence of the ocean, constant wave motion and shadowing effects can makethe process a challenging one.

Two different approaches to object segmentation were carried out; the firstapproach used a background extraction technique to segment foreground objects,and the second approach utilized the benefits of the first technique but subsequentlyapplied an exhaustive search to regions of interest (ROI).

3.1.1. Background extraction

Background extraction is a common method for separating foreground objects frombackground detail in an image scene.18–20 To detect people in beach imagery,background information was extracted using a moving average algorithm (seeFig. 2). The moving average algorithm was originally proposed for highlightingtext in images21 but can also be applied to beach image analysis. The movingaverage algorithm detects the mean gray level of the last n pixels processed, whenconsidering the image as a one-dimensional stream of pixel values [see Eq. (1)].

Mi+1 = Mi − Mi

n+ gi+1. (1)

In Eq. (1), Mi+1 is the estimated moving average for pixel i + 1 having a gray-scale value of gi+1 and Mi is the previous moving average value. If a pixel is lessthan the moving average, it is set to black, otherwise it is set to white. As maybe seen in Fig. 2, the black and white image obtained from the moving averagealgorithm provides an encouraging result. The next stage is to segment the objectsout of the now binary scene.

Int.

J. C

omp.

Int

el. A

ppl.

2006

.06:

149-

160.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

DA

DE

DE

SA

O P

AU

LO

on

05/2

1/13

. For

per

sona

l use

onl

y.

Page 5: 0-0. The Detection of Persons in Cluttered Beach Scenes Using Digital Video Imagery And Neural Network-Based Classification.pdf

March 27, 2007 14:53 WSPC/157-IJCIA 00192

Detection of Persons in Cluttered Beach Scenes 153

Fig. 2. Object detection and segmentation process shown from left to right. The original image,the image after applying the moving average algorithm, segmented images, image segment con-taining the gray-scale image, and the edge detected image (not to scale).

Fig. 3. Object segmentation. The original image is shown on the left, and the segmented objectson the right. The segmentation settings were r = 2 and n = 4.

Objects were segmented out of the binary image using a breadth first search,starting at the top left of the image. Pixels were grouped to form objects based ontheir radius r from other pixels. In Fig. 3, the image contains seven distinct groupsof pixels, with some groups having only one pixel. If the above process is appliedwith a radius r = 2 and a minimum number of pixels n = 4, then only the threeobjects displayed on the right in Fig. 3 will be extracted.

The segmented components are now used to locate the objects in the original(gray-scale) image scene. These objects are subsequently processed by the Cannyedge detection algorithm22 and a feature vector for each object (created using themodified direction feature (MDF) extractor) is presented to a multi-layer perceptron(MLP) for classification.

3.1.2. Exhaustive search

The background extraction technique described above provides a fast methodfor removing background information such as sand, sea and sky. Unfortunately,foreground objects (blobs) detected using this technique can be joined to otherobjects by shadows, the proximity to other objects, and occlusion. An exhaustivesearch method is subsequently put forward to try and overcome the aforementioned

Int.

J. C

omp.

Int

el. A

ppl.

2006

.06:

149-

160.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

DA

DE

DE

SA

O P

AU

LO

on

05/2

1/13

. For

per

sona

l use

onl

y.

Page 6: 0-0. The Detection of Persons in Cluttered Beach Scenes Using Digital Video Imagery And Neural Network-Based Classification.pdf

March 27, 2007 14:53 WSPC/157-IJCIA 00192

154 S. Green et al.

Fig. 4. Exhaustive search window increasing in size to locate person objects.

Fig. 5. Exhaustive search technique locating persons based on foreground objects revealed byapplying background extraction.

problems. The exhaustive search technique described in this research applies asearch window of increasing size over ROIs to detect person and non-person objects(see Figs. 4 and 5).

As the name implies, the exhaustive search method is highly computationallyintensive. To reduce the search time, regions that were not of interest needed tobe excluded quickly. One method to exclude these regions quickly is to providea hierarchy of classifiers, with the simplest and most efficient classifier utilizedfirst to remove non-objects. The proposed object detection system incorporates asingle classifier (MLP). Regions containing non-objects are removed from the searchspace by applying background extraction, and then building a summed-area table(SAT)23 of the binarized image. The SAT can then be used to quickly look at an

Int.

J. C

omp.

Int

el. A

ppl.

2006

.06:

149-

160.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

DA

DE

DE

SA

O P

AU

LO

on

05/2

1/13

. For

per

sona

l use

onl

y.

Page 7: 0-0. The Detection of Persons in Cluttered Beach Scenes Using Digital Video Imagery And Neural Network-Based Classification.pdf

March 27, 2007 14:53 WSPC/157-IJCIA 00192

Detection of Persons in Cluttered Beach Scenes 155

ROI to see if it contains foreground pixels.24 In order to provoke a response, anROI was required to contain between 20% and 80% of foreground pixels comparedwith the maximum possible number of foreground pixels. This was an empiricalmeasurement based on observation. If an ROI is classified as containing a personobject, the region is removed from the search space by setting the foreground pixelsto “white”, and then rebuilding the SAT.

3.2. Modified direction feature

Modified direction feature (MDF) has been detailed elsewhere25 and will only bebriefly described here. MDF feature vector creation is based on the location oftransitions from background to foreground pixels in the vertical and horizontaldirections of an image represented by edges. When a transition is located, twovalues are stored: the location of the transition (LT) and the direction transition(DT) (see Fig. 6). An LT is calculated by taking the ratio between the positionwhere a transition occurs and the distance across the entire image in a particulardirection. The DT value at a particular location is also stored. The DT is calcu-lated by examining the stroke direction of an object’s boundary at the positionwhere a transition occurs (as defined in Blumenstein et al.25) Finally, a vectorcomprising the [LT, DT] values in each of the four possible traversal directionsis created.

Fig. 6. In the figure, the MDF extraction technique examines the person object in the left-to-rightdirection to determine the LT and DT values.

Int.

J. C

omp.

Int

el. A

ppl.

2006

.06:

149-

160.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

DA

DE

DE

SA

O P

AU

LO

on

05/2

1/13

. For

per

sona

l use

onl

y.

Page 8: 0-0. The Detection of Persons in Cluttered Beach Scenes Using Digital Video Imagery And Neural Network-Based Classification.pdf

March 27, 2007 14:53 WSPC/157-IJCIA 00192

156 S. Green et al.

3.3. Neural-based classification

In this research, an interconnected, feed-forward MLP was adopted for the processof classifying objects into person and non-person categories. The MLP was trainedusing the backpropagation (BP) algorithm. In the experiments conducted, the BP-MLP was trained with images processed by the MDF extraction technique.

4. Results

4.1. Image database

To test the person detection system on beach scenes using the MDF feature extrac-tion technique, a database was created including people and non-people objects.The image database consisted of 450 images of varying size. These images wereextracted from DVD recordings taken from a Coastalwatch web camera at SurfersParadise. Frames from the beach recordings were extracted, and image regions wereclassified by a human operator into two categories; “person” and “non-person”.The image database was then broken into three sets to facilitate three-fold cross-validation, whereby each set was composed of 350 training images and 100 testimages. Cross-validation was performed to verify the results obtained for the vari-ous neural network experiments.

4.2. Classifier settings

As previously mentioned, the image database was divided into separate sets toperform three-fold cross-validation. On each set, training was conducted using 8,12, 16, 20, 24, and 28 hidden units, and the RMS error rate was recorded. An RMSerror rate of 0.001 was used as a stopping criterion, but in some cases the MLP didnot train to an RMS of 0.001. The results from all three sets were averaged andcan be seen in Table 1.

4.3. Classification results

In this section, results for the classification of person objects are presented in tabularform. Table 1 presents the results obtained using MDF for processing the test imageset. The MLP parameters for training were modified to optimize the training result.The test set classification rate was calculated based upon the number of successfullyrecognized person objects (true positives). Also listed below is the number of “non-person” items incorrectly labeled as person objects (false positives). The table showsthat, overall, the system provided a good result considering the complexity of thebeach scenes.

4.4. Discussion of classification results

As may be seen from Table 1, the best result was a classification rate of 85% forpeople with a false positive rate of 5% using the MDF technique and an MLP

Int.

J. C

omp.

Int

el. A

ppl.

2006

.06:

149-

160.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

DA

DE

DE

SA

O P

AU

LO

on

05/2

1/13

. For

per

sona

l use

onl

y.

Page 9: 0-0. The Detection of Persons in Cluttered Beach Scenes Using Digital Video Imagery And Neural Network-Based Classification.pdf

March 27, 2007 14:53 WSPC/157-IJCIA 00192

Detection of Persons in Cluttered Beach Scenes 157

Table 1. BP-MLP classification rates using the MDF extraction technique

Person Object Classification Rate (%)

No. of Hidden Units RMS Error True Positive False Positive

8 0.01375233 87 1112 0.01390490 88 916 0.01404126 85 720 0.01414976 85 624 0.03212585 86 528 0.01438113 85 5

Fig. 7. Person detection results showing false positives (top row) and true positives (bottomrow); both sets of images are not to scale.

with 28 hidden units. Figure 7 shows some results for cases of false positives andtrue positives. By observation of the data, people who were not identified correctlywere either lying on the beach, occluded by other objects (both person and non-person), or were obscured by the surf. In the case of false positives, objects thatwere classified as a person when the object was a non-person, tended to have anoutline that was upright, causing the MLP to give an incorrect classification.

4.5. Comparison between background extraction and exhaustive

search results

Experiments were carried out to compare both background extraction and exhaus-tive search techniques. The experiments were conducted on 10 minutes of beachfootage recorded from a Coastalwatch web camera located at Surfers Paradise beachin Queensland, Australia. The beach scene in all frames contained approximately31 people. For both techniques, each frame was recorded with white rectangles rep-resenting objects classified as a person (true positive), and black rectangles repre-senting objects classified as non-person (false positive). In the case of the exhaustivesearch, black regions were not shown due to the large number of these generatedby the exhaustive search method (see Fig. 8). In Table 2, it can be seen that thebackground extraction technique did not deal with occlusion and shadowing effectsvery well; however, it did manage to locate most areas containing person objects.The background extraction technique had a very low false positive rate in thatthere were very few non-person objects classified incorrectly. The exhaustive searchmethod gave a much better result for locating person objects that were occludedby other objects and were obscured by shadowing effects, with a best result of

Int.

J. C

omp.

Int

el. A

ppl.

2006

.06:

149-

160.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

DA

DE

DE

SA

O P

AU

LO

on

05/2

1/13

. For

per

sona

l use

onl

y.

Page 10: 0-0. The Detection of Persons in Cluttered Beach Scenes Using Digital Video Imagery And Neural Network-Based Classification.pdf

March 27, 2007 14:53 WSPC/157-IJCIA 00192

158 S. Green et al.

Fig. 8. Persons located in beach imagery using background extraction and exhaustive searchmethods.

Table 2. Results for both background extraction and exhaustivesearch methods, showing the number of people detected (truepositive) in each frame (average of 31 people in each frame).

Frame No. True Positive False Positive

Background extraction10010 5 110080 8 110110 5 2

Exhaustive Search10010 19 410080 22 410110 20 10

22 persons correctly located. The disadvantage of the exhaustive search methodwas the substantial increase in false positive classifications.

5. Conclusions and Future Research

In this paper, a person detection system for quantifying people in beach scenes hasbeen described. A human operator segmented actual beach data from a Surfers Par-adise web camera into person and non-person categories. Tests were then carried outon images using two techniques: (1) background extraction (2) and the exhaustive

Int.

J. C

omp.

Int

el. A

ppl.

2006

.06:

149-

160.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

DA

DE

DE

SA

O P

AU

LO

on

05/2

1/13

. For

per

sona

l use

onl

y.

Page 11: 0-0. The Detection of Persons in Cluttered Beach Scenes Using Digital Video Imagery And Neural Network-Based Classification.pdf

March 27, 2007 14:53 WSPC/157-IJCIA 00192

Detection of Persons in Cluttered Beach Scenes 159

search technique using beach footage. The system incorporated a neural-basedclassification technique for the detection of person and non-person objects. Encour-aging results were presented for automated person detection and quantification,which can be applied to beach scene analysis for surveillance and coastal manage-ment applications.

Currently, the features described in this paper are not invariant to object rota-tion. Future research will be directed to address this point. Further work will alsoexamine different object segmentation techniques, and determine ways to furtherdeal with images that are not correctly classified. Finally, frames retrieved frombeach media are at present processed individually. Future research will look at gen-erating a confidence value for a person object based on temporal data from pastframes. Also, motion detection may provide a more effective way of detecting personobjects in beach imagery.

Acknowledgments

The authors acknowledge the assistance of Chris Lane of CoastalWatch for assis-tance in providing video imagery for this research.

References

1. M. Browne, M. Blumenstein, R. Tomlinson and C. Lane, An intelligent system forremote monitoring and prediction of beach conditions, Proc. Int. Conf. Artif. Intell.Appl., Innsbruck (2005), pp. 533–537.

2. A. J. Schofield, T. J. Stonham and P. A. Mehta, Automated people counting to aidlift control, Automat. Constr. 6(5–6) (1997) 437–445.

3. A. J. Schofield, P. A. Mehta and T. J. Stonham, A system for counting people in videoimages using neural networks to identify the background scene, Pattern Recogn. 29(8)(1996) 1421–1428.

4. A. Iketani, A. Nagai, Y. Kuno and Y. Shirai, Real-time surveillance system detectingpersons in complex scenes, Real-Time Imaging 7(5) (2001) 433–446.

5. C. Sacchi, G. Gera, L. Marcenaro and C. S. Regazzoni, Advanced image-processingtools for counting people in tourist site-monitoring applications, Signal Process. 81(5)(2001) 1017–1040.

6. F. Bartolini, V. Cappellini and A. Mecocci, Counting people getting in and out ofa bus by real-time image sequence processing, Image Vision Comput. 12(1) (1994)36–41.

7. C.-J. Pai, H.-R. Tyan, Y.-M. Liang, H.-Y. M. Liao and S.-W. Chen, Pedestrian detec-tion and tracking at crossroads, Pattern Recogn. 37(5) (2004) 1025–1034.

8. G. L. Foresti, C. Micheloni and C. Piciarelli, Detecting moving people in videostreams, Pattern Recogn. 26(14) (2005) 2232–2243.

9. Y. Ricquebourg and P. Bouthemy, Real-time tracking of moving persons by exploitingspatio-temporal image slices, IEEE Trans. Pattern Anal. Mach. Intell. 22(8) (2000)797–808.

10. C. Lerdsudwichai, M. Abdel-Mottaleb and A.-N. Ansari, Tracking multiple peoplewith recovery from partial and total occlusion, Pattern Recogn. 38(7) (2005) 1059–1070.

Int.

J. C

omp.

Int

el. A

ppl.

2006

.06:

149-

160.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

DA

DE

DE

SA

O P

AU

LO

on

05/2

1/13

. For

per

sona

l use

onl

y.

Page 12: 0-0. The Detection of Persons in Cluttered Beach Scenes Using Digital Video Imagery And Neural Network-Based Classification.pdf

March 27, 2007 14:53 WSPC/157-IJCIA 00192

160 S. Green et al.

11. J.-C. Tai, S.-T. Tseng, C.-P. Lin and K.-T. Song, Real-time image tracking for auto-matic traffic monitoring and enforcement applications, Image Vision Comput. 22(6)(2004) 485–501.

12. D. M. Ha, J.-M. Lee, and Y.-D. Kim, Neural-edge-based vehicle detection and trafficparameter extraction, Image Vision Comput. 22 (2004) 899–907.

13. C. Wohler and J. K. Anlauf, Real-time object recognition on image sequences withthe adaptable time delay neural network algorithm — Applications for autonomousvehicles, Image Vision Comput. 19(9–10) (2001) 593–618.

14. J. Altmann, S. Linev and A. Wei, Acoustic-seismic detection and classification of mili-tary vehicles — Developing tools for disarmament and peace-keeping, Appl. Acoustics63(10) (2002) 1085–1107.

15. H. Zhang, W. Gao and D. Zhao, Object Detection Using Spatial Histogram Features,Image Vision Comput. 24(4) (2006) 327–341.

16. H. Schneiderman, Feature-centric evaluation for efficient cascaded object detection,Conf. Comput. Vision Pattern Recogn. (CVPR), 2 (2004) 29–36.

17. D. Hoiem, R. Sukthankar, H. Schneiderman and L. Huston, Object-based imageretrieval using the statistical structure of images, Conf. Comput. Vision PatternRecogn. (CVPR) 2 (2004) 490–497.

18. P. Figueroa, N. Leite and R. Barros, Background recovering in outdoor imagesequences, Image Vision Comput. 24(4) (2006) 363–374.

19. B. Shoushtarian and H. E. Bez, A practical adaptive approach for dynamic back-ground subtraction using an invariant colour model and object tracking, PatternRecogn. Lett. 26(1) (2005) 5–26.

20. I. Haritaoglu, D. Harwood and L. S. Davis, W4: Real-time surveillance of people andtheir activities, IEEE Trans. Pattern Anal. Mach. Int. 22(8) (2000) 809–830.

21. J. R. Parker. Algorithms for Image Processing and Computer Vision (John Wiley andSons, New York, 1997).

22. J. F. Canny, Computational approach to edge detection. IEEE Trans. Pattern Anal.Mach. Intell. 8(6) (1986) 679–698.

23. F. Crow, Summed-area tables for texture mapping. Comput. Graphics (SIGGRAPH’84 Proceedings) 18(3) (1984) 207–212.

24. P. Viola and M. Jones, Robust real-time object detection, in 2nd International Work-shop Statistical and Computational Theories of Vision — Modeling, Learning, Com-puting, and Sampling, Vancouver, Canada (2001).

25. M. Blumenstein, X. Y. Liu and B. Verma, A modified direction feature for cur-sive character recognition, in Proc. Int. Joint Conf. Neural Networks (IJCNN ’04),Budapest, Hungary (2004), pp. 2983–2987.In

t. J.

Com

p. I

ntel

. App

l. 20

06.0

6:14

9-16

0. D

ownl

oade

d fr

om w

ww

.wor

ldsc

ient

ific

.com

by U

NIV

ER

SID

AD

E D

E S

AO

PA

UL

O o

n 05

/21/

13. F

or p

erso

nal u

se o

nly.


Related Documents