Radar image based road extraction and vehicle detection for autonomous surveillance systems

Erfu Yang, Fei Gao, Wenguang Wang, Amir Hussain, Jun Wang, Bin Luo

Research output: Contribution to conferenceSpeech

Abstract

Visual surveillance is an attempt to detect, recognize and track certain objects from image sequences, and more generally to understand and describe object behaviors. Future’s visual surveillance systems for outdoor (including security) applications will require mission platforms that are autonomous, asynchronous, adaptive and highly sensitive in complex, time-varying and possibly hostile environments.
One fundamental problem in autonomous surveillance systems is how to detect multiple objects, sense and percept the world/environment in extreme outdoor conditions in which the presence of dust, fog, rain, changing illumination can dramatically degrade conventional stereo and laser sensing. Thus, radar-based imaging in autonomous surveillance systems has attracted extensive research attention in recent years since radar also allows for multiple object detection within a single beam, whereas other range sensors are limited to one target return per emission. The main challenge arising from radar-based autonomous visual surveillance systems is always linked to radar image information processing and utilization, i.e., how to quickly and efficiently extract and analyse the information of interest for multiple object detection and extraction such as road networks, detecting vehicles from consecutive images acquired by radar imaging sensors.
In this talk, we firstly present a novel neurobiologically-inspired approach to road extraction from radar images. The proposed approach is based on the neurobiological mechanism in which simple cells in primary visual cortex are believed to extract local contour information from a visual scene. The CORF (Combination of Receptive Fields) computational model is used in the proposed approach to road extraction from radar images. To demonstrate the effectiveness of the proposed approach, the numerical experimental results obtained from using several SAR(synthetic aperture radar) images are provided. Secondly, a visual attention model-based algorithm is proposed to detect vehicles from SAR (synthetic aperture radar) images. The performance of the proposed algorithm is demonstrated by using real SAR images with 20 vehicle targets. Finally, an algorithm based on kernel fisher discriminant analysis (KFDA) for vehicle detection and recognition in SAR images is presented. We obtain image samples with a dual-window approach and extract features of the inner and outer window samples using KFDA. An improved KFDA-IMED (Image Euclidean Distance) strategy is employed to obtain the projection in known sample space and combine with a support vector machine (SVM) to recognize vehicles effectively. The performance of this new method is validated with the MSTAR (moving and stationary target acquisition and recognition) database.
Acknowledgements:
This research is supported by The Royal Society of Edinburgh (RSE) and The National Natural Science Foundation of China (NNSFC) under the RSE-NNSFC joint projects (2012-2014) [grant number 61211130309 and 61211130210] with Beihang University and Anhui University, China, respectively. It is supported in part by the “Sino-UK Higher Education Research Partnership for PhD Studies” joint-project (2013-2015) funded by the British Council China and The China Scholarship Council (CSC). Both Amir Hussain and Erfu Yang were also funded, in part, by the UK Engineering and Physical Sciences Research Council (EPSRC) [grant number EP/I009310/1].
Original languageEnglish
Pages8-9
Number of pages2
Publication statusPublished - 16 Oct 2014
EventFifth China-Scotland SIPRA Workshop on Recent Advances in Signal and Image Processing - ANHUI UNIVERSITY, Hefei, China
Duration: 16 Oct 201518 Oct 2015

Workshop

WorkshopFifth China-Scotland SIPRA Workshop on Recent Advances in Signal and Image Processing
CountryChina
CityHefei
Period16/10/1518/10/15

Fingerprint

Radar
Synthetic aperture radar
Discriminant analysis
Natural sciences
Radar imaging
Sensors
Fog
Support vector machines
Rain
Dust
Lighting
Education
Imaging techniques
Lasers
Object detection

Keywords

  • image processing
  • target detection
  • autonomous systems
  • cognitive computation
  • visual attention model
  • computer vision

Cite this

Yang, E., Gao, F., Wang, W., Hussain, A., Wang, J., & Luo, B. (2014). Radar image based road extraction and vehicle detection for autonomous surveillance systems. 8-9. Fifth China-Scotland SIPRA Workshop on Recent Advances in Signal and Image Processing, Hefei, China.
Yang, Erfu ; Gao, Fei ; Wang, Wenguang ; Hussain, Amir ; Wang, Jun ; Luo, Bin. / Radar image based road extraction and vehicle detection for autonomous surveillance systems. Fifth China-Scotland SIPRA Workshop on Recent Advances in Signal and Image Processing, Hefei, China.2 p.
@conference{c1b56becd8f34abd80e44d05fef29d9b,
title = "Radar image based road extraction and vehicle detection for autonomous surveillance systems",
abstract = "Visual surveillance is an attempt to detect, recognize and track certain objects from image sequences, and more generally to understand and describe object behaviors. Future’s visual surveillance systems for outdoor (including security) applications will require mission platforms that are autonomous, asynchronous, adaptive and highly sensitive in complex, time-varying and possibly hostile environments.One fundamental problem in autonomous surveillance systems is how to detect multiple objects, sense and percept the world/environment in extreme outdoor conditions in which the presence of dust, fog, rain, changing illumination can dramatically degrade conventional stereo and laser sensing. Thus, radar-based imaging in autonomous surveillance systems has attracted extensive research attention in recent years since radar also allows for multiple object detection within a single beam, whereas other range sensors are limited to one target return per emission. The main challenge arising from radar-based autonomous visual surveillance systems is always linked to radar image information processing and utilization, i.e., how to quickly and efficiently extract and analyse the information of interest for multiple object detection and extraction such as road networks, detecting vehicles from consecutive images acquired by radar imaging sensors. In this talk, we firstly present a novel neurobiologically-inspired approach to road extraction from radar images. The proposed approach is based on the neurobiological mechanism in which simple cells in primary visual cortex are believed to extract local contour information from a visual scene. The CORF (Combination of Receptive Fields) computational model is used in the proposed approach to road extraction from radar images. To demonstrate the effectiveness of the proposed approach, the numerical experimental results obtained from using several SAR(synthetic aperture radar) images are provided. Secondly, a visual attention model-based algorithm is proposed to detect vehicles from SAR (synthetic aperture radar) images. The performance of the proposed algorithm is demonstrated by using real SAR images with 20 vehicle targets. Finally, an algorithm based on kernel fisher discriminant analysis (KFDA) for vehicle detection and recognition in SAR images is presented. We obtain image samples with a dual-window approach and extract features of the inner and outer window samples using KFDA. An improved KFDA-IMED (Image Euclidean Distance) strategy is employed to obtain the projection in known sample space and combine with a support vector machine (SVM) to recognize vehicles effectively. The performance of this new method is validated with the MSTAR (moving and stationary target acquisition and recognition) database.Acknowledgements: This research is supported by The Royal Society of Edinburgh (RSE) and The National Natural Science Foundation of China (NNSFC) under the RSE-NNSFC joint projects (2012-2014) [grant number 61211130309 and 61211130210] with Beihang University and Anhui University, China, respectively. It is supported in part by the “Sino-UK Higher Education Research Partnership for PhD Studies” joint-project (2013-2015) funded by the British Council China and The China Scholarship Council (CSC). Both Amir Hussain and Erfu Yang were also funded, in part, by the UK Engineering and Physical Sciences Research Council (EPSRC) [grant number EP/I009310/1].",
keywords = "image processing, target detection, autonomous systems, cognitive computation, visual attention model , computer vision",
author = "Erfu Yang and Fei Gao and Wenguang Wang and Amir Hussain and Jun Wang and Bin Luo",
year = "2014",
month = "10",
day = "16",
language = "English",
pages = "8--9",
note = "Fifth China-Scotland SIPRA Workshop on Recent Advances in Signal and Image Processing ; Conference date: 16-10-2015 Through 18-10-2015",

}

Yang, E, Gao, F, Wang, W, Hussain, A, Wang, J & Luo, B 2014, 'Radar image based road extraction and vehicle detection for autonomous surveillance systems' Fifth China-Scotland SIPRA Workshop on Recent Advances in Signal and Image Processing, Hefei, China, 16/10/15 - 18/10/15, pp. 8-9.

Radar image based road extraction and vehicle detection for autonomous surveillance systems. / Yang, Erfu; Gao, Fei; Wang, Wenguang; Hussain, Amir; Wang, Jun; Luo, Bin.

2014. 8-9 Fifth China-Scotland SIPRA Workshop on Recent Advances in Signal and Image Processing, Hefei, China.

Research output: Contribution to conferenceSpeech

TY - CONF

T1 - Radar image based road extraction and vehicle detection for autonomous surveillance systems

AU - Yang, Erfu

AU - Gao, Fei

AU - Wang, Wenguang

AU - Hussain, Amir

AU - Wang, Jun

AU - Luo, Bin

PY - 2014/10/16

Y1 - 2014/10/16

N2 - Visual surveillance is an attempt to detect, recognize and track certain objects from image sequences, and more generally to understand and describe object behaviors. Future’s visual surveillance systems for outdoor (including security) applications will require mission platforms that are autonomous, asynchronous, adaptive and highly sensitive in complex, time-varying and possibly hostile environments.One fundamental problem in autonomous surveillance systems is how to detect multiple objects, sense and percept the world/environment in extreme outdoor conditions in which the presence of dust, fog, rain, changing illumination can dramatically degrade conventional stereo and laser sensing. Thus, radar-based imaging in autonomous surveillance systems has attracted extensive research attention in recent years since radar also allows for multiple object detection within a single beam, whereas other range sensors are limited to one target return per emission. The main challenge arising from radar-based autonomous visual surveillance systems is always linked to radar image information processing and utilization, i.e., how to quickly and efficiently extract and analyse the information of interest for multiple object detection and extraction such as road networks, detecting vehicles from consecutive images acquired by radar imaging sensors. In this talk, we firstly present a novel neurobiologically-inspired approach to road extraction from radar images. The proposed approach is based on the neurobiological mechanism in which simple cells in primary visual cortex are believed to extract local contour information from a visual scene. The CORF (Combination of Receptive Fields) computational model is used in the proposed approach to road extraction from radar images. To demonstrate the effectiveness of the proposed approach, the numerical experimental results obtained from using several SAR(synthetic aperture radar) images are provided. Secondly, a visual attention model-based algorithm is proposed to detect vehicles from SAR (synthetic aperture radar) images. The performance of the proposed algorithm is demonstrated by using real SAR images with 20 vehicle targets. Finally, an algorithm based on kernel fisher discriminant analysis (KFDA) for vehicle detection and recognition in SAR images is presented. We obtain image samples with a dual-window approach and extract features of the inner and outer window samples using KFDA. An improved KFDA-IMED (Image Euclidean Distance) strategy is employed to obtain the projection in known sample space and combine with a support vector machine (SVM) to recognize vehicles effectively. The performance of this new method is validated with the MSTAR (moving and stationary target acquisition and recognition) database.Acknowledgements: This research is supported by The Royal Society of Edinburgh (RSE) and The National Natural Science Foundation of China (NNSFC) under the RSE-NNSFC joint projects (2012-2014) [grant number 61211130309 and 61211130210] with Beihang University and Anhui University, China, respectively. It is supported in part by the “Sino-UK Higher Education Research Partnership for PhD Studies” joint-project (2013-2015) funded by the British Council China and The China Scholarship Council (CSC). Both Amir Hussain and Erfu Yang were also funded, in part, by the UK Engineering and Physical Sciences Research Council (EPSRC) [grant number EP/I009310/1].

AB - Visual surveillance is an attempt to detect, recognize and track certain objects from image sequences, and more generally to understand and describe object behaviors. Future’s visual surveillance systems for outdoor (including security) applications will require mission platforms that are autonomous, asynchronous, adaptive and highly sensitive in complex, time-varying and possibly hostile environments.One fundamental problem in autonomous surveillance systems is how to detect multiple objects, sense and percept the world/environment in extreme outdoor conditions in which the presence of dust, fog, rain, changing illumination can dramatically degrade conventional stereo and laser sensing. Thus, radar-based imaging in autonomous surveillance systems has attracted extensive research attention in recent years since radar also allows for multiple object detection within a single beam, whereas other range sensors are limited to one target return per emission. The main challenge arising from radar-based autonomous visual surveillance systems is always linked to radar image information processing and utilization, i.e., how to quickly and efficiently extract and analyse the information of interest for multiple object detection and extraction such as road networks, detecting vehicles from consecutive images acquired by radar imaging sensors. In this talk, we firstly present a novel neurobiologically-inspired approach to road extraction from radar images. The proposed approach is based on the neurobiological mechanism in which simple cells in primary visual cortex are believed to extract local contour information from a visual scene. The CORF (Combination of Receptive Fields) computational model is used in the proposed approach to road extraction from radar images. To demonstrate the effectiveness of the proposed approach, the numerical experimental results obtained from using several SAR(synthetic aperture radar) images are provided. Secondly, a visual attention model-based algorithm is proposed to detect vehicles from SAR (synthetic aperture radar) images. The performance of the proposed algorithm is demonstrated by using real SAR images with 20 vehicle targets. Finally, an algorithm based on kernel fisher discriminant analysis (KFDA) for vehicle detection and recognition in SAR images is presented. We obtain image samples with a dual-window approach and extract features of the inner and outer window samples using KFDA. An improved KFDA-IMED (Image Euclidean Distance) strategy is employed to obtain the projection in known sample space and combine with a support vector machine (SVM) to recognize vehicles effectively. The performance of this new method is validated with the MSTAR (moving and stationary target acquisition and recognition) database.Acknowledgements: This research is supported by The Royal Society of Edinburgh (RSE) and The National Natural Science Foundation of China (NNSFC) under the RSE-NNSFC joint projects (2012-2014) [grant number 61211130309 and 61211130210] with Beihang University and Anhui University, China, respectively. It is supported in part by the “Sino-UK Higher Education Research Partnership for PhD Studies” joint-project (2013-2015) funded by the British Council China and The China Scholarship Council (CSC). Both Amir Hussain and Erfu Yang were also funded, in part, by the UK Engineering and Physical Sciences Research Council (EPSRC) [grant number EP/I009310/1].

KW - image processing

KW - target detection

KW - autonomous systems

KW - cognitive computation

KW - visual attention model

KW - computer vision

M3 - Speech

SP - 8

EP - 9

ER -

Yang E, Gao F, Wang W, Hussain A, Wang J, Luo B. Radar image based road extraction and vehicle detection for autonomous surveillance systems. 2014. Fifth China-Scotland SIPRA Workshop on Recent Advances in Signal and Image Processing, Hefei, China.