The Experts below are selected from a list of 132 Experts worldwide ranked by ideXlab platform

Zhicheng Zhao - One of the best experts on this subject based on the ideXlab platform.

  • Unsupervised Anomaly Detection for Traffic Surveillance Based on Background Modeling
    2018 IEEE CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2018
    Co-Authors: Jiayi Wei, Jianfei Zhao, Yanyun Zhao, Zhicheng Zhao
    Abstract:

    Most state-of-the-art Anomaly detection methods are specific to Detecting Anomaly for pedestrians and cannot work without adequate normal training videos. Recently, there is a growing demand for Detecting anomalous vehicles in traffic surveillance videos. However, the biggest challenge in this task is the lack of labeled datasets for training supervised models. By examining the resemblances of anomalous vehicles, we find it reasonable to label a vehicle as Anomaly if it stays still in the video for a relatively long time. Utilizing this property, in this paper we introduce a novel unsupervised Anomaly detection method for traffic surveillance based on background modeling, which shows great potentials in handling heterogeneous scenes as well as extremely low resolution videos recordings without the dependence on labeled data. In the proposed system, we first employ background modeling using MOG2 to remove the moving vehicles as foreground while keeping the stopped vehicles as part of the background. Then we use Faster R-CNN to detect vehicles in the extracted background and decide if they are new anomalies under certain conditions. All information is updated on a frame basis until the end of the video which contains the final results. In this way, we make full use of the characteristics that abnormal vehicles stay in the scene for a relatively long time and reduce the difficulty of vehicle Anomaly detection. Eventually, we can detect almost every Anomaly in the NVIDIA AI CITY CHALLENGE track-2 dataset except for several extremely complex cases with a 81.08% F1-score and 10.2369 RMSE.

  • CVPR Workshops - Unsupervised Anomaly Detection for Traffic Surveillance Based on Background Modeling
    2018 IEEE CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2018
    Co-Authors: Jiayi Wei, Jianfei Zhao, Zhao Yanyun, Zhicheng Zhao
    Abstract:

    Most state-of-the-art Anomaly detection methods are specific to Detecting Anomaly for pedestrians and cannot work without adequate normal training videos. Recently, there is a growing demand for Detecting anomalous vehicles in traffic surveillance videos. However, the biggest challenge in this task is the lack of labeled datasets for training supervised models. By examining the resemblances of anomalous vehicles, we find it reasonable to label a vehicle as Anomaly if it stays still in the video for a relatively long time. Utilizing this property, in this paper we introduce a novel unsupervised Anomaly detection method for traffic surveilliance based on background modeling, which shows great potentials in handling heterogeneous scenes as well as extremely low resolution videos recordings without the dependence on labeled data. In the proposed system, we first employ background modeling using MOG2 to remove the moving vehicles as foreground while keeping the stopped vehicles as part of the background. Then we use Faster R-CNN to detect vehicles in the extracted background and decide if they are new anomalies under certain conditions. All information is updated on a frame basis until the end of the video which contains the final results. In this way, we make full use of the characteristics that abnormal vehicles stay in the scene for a relatively long time and reduce the difficulty of vehicle Anomaly detection. Eventually, we can detect almost every Anomaly in the NVIDIA AI CITY CHALLENGE track-2 dataset except for several extremely complex cases with a 81.08% F1-score and 10.2369 RMSE.

Jiayi Wei - One of the best experts on this subject based on the ideXlab platform.

  • Unsupervised Anomaly Detection for Traffic Surveillance Based on Background Modeling
    2018 IEEE CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2018
    Co-Authors: Jiayi Wei, Jianfei Zhao, Yanyun Zhao, Zhicheng Zhao
    Abstract:

    Most state-of-the-art Anomaly detection methods are specific to Detecting Anomaly for pedestrians and cannot work without adequate normal training videos. Recently, there is a growing demand for Detecting anomalous vehicles in traffic surveillance videos. However, the biggest challenge in this task is the lack of labeled datasets for training supervised models. By examining the resemblances of anomalous vehicles, we find it reasonable to label a vehicle as Anomaly if it stays still in the video for a relatively long time. Utilizing this property, in this paper we introduce a novel unsupervised Anomaly detection method for traffic surveillance based on background modeling, which shows great potentials in handling heterogeneous scenes as well as extremely low resolution videos recordings without the dependence on labeled data. In the proposed system, we first employ background modeling using MOG2 to remove the moving vehicles as foreground while keeping the stopped vehicles as part of the background. Then we use Faster R-CNN to detect vehicles in the extracted background and decide if they are new anomalies under certain conditions. All information is updated on a frame basis until the end of the video which contains the final results. In this way, we make full use of the characteristics that abnormal vehicles stay in the scene for a relatively long time and reduce the difficulty of vehicle Anomaly detection. Eventually, we can detect almost every Anomaly in the NVIDIA AI CITY CHALLENGE track-2 dataset except for several extremely complex cases with a 81.08% F1-score and 10.2369 RMSE.

  • CVPR Workshops - Unsupervised Anomaly Detection for Traffic Surveillance Based on Background Modeling
    2018 IEEE CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2018
    Co-Authors: Jiayi Wei, Jianfei Zhao, Zhao Yanyun, Zhicheng Zhao
    Abstract:

    Most state-of-the-art Anomaly detection methods are specific to Detecting Anomaly for pedestrians and cannot work without adequate normal training videos. Recently, there is a growing demand for Detecting anomalous vehicles in traffic surveillance videos. However, the biggest challenge in this task is the lack of labeled datasets for training supervised models. By examining the resemblances of anomalous vehicles, we find it reasonable to label a vehicle as Anomaly if it stays still in the video for a relatively long time. Utilizing this property, in this paper we introduce a novel unsupervised Anomaly detection method for traffic surveilliance based on background modeling, which shows great potentials in handling heterogeneous scenes as well as extremely low resolution videos recordings without the dependence on labeled data. In the proposed system, we first employ background modeling using MOG2 to remove the moving vehicles as foreground while keeping the stopped vehicles as part of the background. Then we use Faster R-CNN to detect vehicles in the extracted background and decide if they are new anomalies under certain conditions. All information is updated on a frame basis until the end of the video which contains the final results. In this way, we make full use of the characteristics that abnormal vehicles stay in the scene for a relatively long time and reduce the difficulty of vehicle Anomaly detection. Eventually, we can detect almost every Anomaly in the NVIDIA AI CITY CHALLENGE track-2 dataset except for several extremely complex cases with a 81.08% F1-score and 10.2369 RMSE.

Tsung-yi Ho - One of the best experts on this subject based on the ideXlab platform.

  • Leveraging Strategic Detection Techniques for Smart Home Pricing Cyberattacks
    IEEE Transactions on Dependable and Secure Computing, 2016
    Co-Authors: Shiyan Hu, Tsung-yi Ho
    Abstract:

    In this work, the vulnerability of the electricity pricing model in the smart home system is assessed. Two closely related pricing cyberattacks which manipulate the guideline electricity prices received at smart meters are considered and they aim at reducing the expense of the cyberattacker and increasing the peak energy usage in the local community. A single event detection technique which uses support vector regression and impact difference for Detecting Anomaly pricing is proposed. The detection capability of such a technique is still limited since it does not model the long term impact of pricing cyberattacks. This motivates us to develop a partially observable Markov decision process based detection algorithm, which has the ingredients such as reward expectation and policy transfer graph to account for the cumulative impact and the potential future impact due to pricing cyberattacks. Our simulation results demonstrate that the pricing cyberattack can reduce the cyberattacker's bill by 34.3 percent at cost of the increase of others’ bill by 7.9 percent, and increase the peak to average ratio (PAR) by 35.7 percent. Furthermore, the proposed long term detection technique has the detection accuracy of more than 97 percent with significant reduction in PAR and bill compared to repeatedly using the single event detection technique.

  • Vulnerability assessment and defense technology for smart home cybersecurity considering pricing cyberattacks
    IEEE ACM International Conference on Computer-Aided Design Digest of Technical Papers ICCAD, 2015
    Co-Authors: Shiyan Hu, Tsung-yi Ho
    Abstract:

    Smart home, which controls the end use of the power grid, has become a critical component in the smart grid infrastructure. In a smart home system, the advanced metering infrastructure (AMI) is used to connect smart meters with the power system and the communication system of a smart grid. The electricity pricing information is transmitted from the utility to the local community, and then broadcast through wired or wireless networks to each smart meter within AMI. In this work, the vulnerability of the above process is assessed. Two closely related pricing cyberattacks which manipulate the guideline electricity prices received at smart meters are considered and they aim at reducing the expense of the cyberattacker and increasing the peak energy usage in the local community. A countermeasure technique which uses support vector regression and impact difference for Detecting Anomaly pricing is then proposed. These pricing cyberattacks explore the interdependance between the transmitted electricity pricing in the communication system and the energy load in the power system, which are the first such cyber-attacks in the smart home context. Our simulation results demonstrate that the pricing cyberattack can reduce the attacker's bill by 34.3% at the cost of the increase of others' bill by 7.9% on average. In addition, the pricing cyberattack can unbalance the energy load of the local power system as it increases the peak to average ratio by 35.7%. Furthermore, our simulation results show that the proposed countermeasure technique can effectively detect the electricity pricing manipulation.

Jianfei Zhao - One of the best experts on this subject based on the ideXlab platform.

  • Unsupervised Anomaly Detection for Traffic Surveillance Based on Background Modeling
    2018 IEEE CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2018
    Co-Authors: Jiayi Wei, Jianfei Zhao, Yanyun Zhao, Zhicheng Zhao
    Abstract:

    Most state-of-the-art Anomaly detection methods are specific to Detecting Anomaly for pedestrians and cannot work without adequate normal training videos. Recently, there is a growing demand for Detecting anomalous vehicles in traffic surveillance videos. However, the biggest challenge in this task is the lack of labeled datasets for training supervised models. By examining the resemblances of anomalous vehicles, we find it reasonable to label a vehicle as Anomaly if it stays still in the video for a relatively long time. Utilizing this property, in this paper we introduce a novel unsupervised Anomaly detection method for traffic surveillance based on background modeling, which shows great potentials in handling heterogeneous scenes as well as extremely low resolution videos recordings without the dependence on labeled data. In the proposed system, we first employ background modeling using MOG2 to remove the moving vehicles as foreground while keeping the stopped vehicles as part of the background. Then we use Faster R-CNN to detect vehicles in the extracted background and decide if they are new anomalies under certain conditions. All information is updated on a frame basis until the end of the video which contains the final results. In this way, we make full use of the characteristics that abnormal vehicles stay in the scene for a relatively long time and reduce the difficulty of vehicle Anomaly detection. Eventually, we can detect almost every Anomaly in the NVIDIA AI CITY CHALLENGE track-2 dataset except for several extremely complex cases with a 81.08% F1-score and 10.2369 RMSE.

  • CVPR Workshops - Unsupervised Anomaly Detection for Traffic Surveillance Based on Background Modeling
    2018 IEEE CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2018
    Co-Authors: Jiayi Wei, Jianfei Zhao, Zhao Yanyun, Zhicheng Zhao
    Abstract:

    Most state-of-the-art Anomaly detection methods are specific to Detecting Anomaly for pedestrians and cannot work without adequate normal training videos. Recently, there is a growing demand for Detecting anomalous vehicles in traffic surveillance videos. However, the biggest challenge in this task is the lack of labeled datasets for training supervised models. By examining the resemblances of anomalous vehicles, we find it reasonable to label a vehicle as Anomaly if it stays still in the video for a relatively long time. Utilizing this property, in this paper we introduce a novel unsupervised Anomaly detection method for traffic surveilliance based on background modeling, which shows great potentials in handling heterogeneous scenes as well as extremely low resolution videos recordings without the dependence on labeled data. In the proposed system, we first employ background modeling using MOG2 to remove the moving vehicles as foreground while keeping the stopped vehicles as part of the background. Then we use Faster R-CNN to detect vehicles in the extracted background and decide if they are new anomalies under certain conditions. All information is updated on a frame basis until the end of the video which contains the final results. In this way, we make full use of the characteristics that abnormal vehicles stay in the scene for a relatively long time and reduce the difficulty of vehicle Anomaly detection. Eventually, we can detect almost every Anomaly in the NVIDIA AI CITY CHALLENGE track-2 dataset except for several extremely complex cases with a 81.08% F1-score and 10.2369 RMSE.

Sarp Ertürk - One of the best experts on this subject based on the ideXlab platform.

  • Anomaly and homogeneous region guided endmember extraction for hyperspectral images
    2013 21st Signal Processing and Communications Applications Conference (SIU), 2013
    Co-Authors: Alp Ertürk, Davut Çeşmeci, Deniz Gerçek, Mehmet Kemal Güllü, Sarp Ertürk
    Abstract:

    In the cases that the spatial resolution of the hyperspectral data is not sufficient, pixel vectors are expressed in terms of abundances of pure signatures, named as endmembers, with spectral mixture analysis. Most of the endmember extraction methods use only the spectral information, whereas spatial pre-processing methods can increase the performance by directing the endmember extraction process to spatially homogeneous regions. However, this approach results in a failure in Detecting Anomaly endmembers. In this paper, a two-way approach which provides high performance by directing the endmember extraction process to both anomalies and spatially homogenous regions is proposed.

  • SIU - Anomaly and homogeneous region guided endmember extraction for hyperspectral images
    2013 21st Signal Processing and Communications Applications Conference (SIU), 2013
    Co-Authors: Alp Ertürk, Davut Çeşmeci, Deniz Gerçek, Mehmet Kemal Güllü, Sarp Ertürk
    Abstract:

    In the cases that the spatial resolution of the hyperspectral data is not sufficient, pixel vectors are expressed in terms of abundances of pure signatures, named as endmembers, with spectral mixture analysis. Most of the endmember extraction methods use only the spectral information, whereas spatial pre-processing methods can increase the performance by directing the endmember extraction process to spatially homogeneous regions. However, this approach results in a failure in Detecting Anomaly endmembers. In this paper, a two-way approach which provides high performance by directing the endmember extraction process to both anomalies and spatially homogenous regions is proposed.