Limits...
Traffic Behavior Recognition Using the Pachinko Allocation Model.

Huynh-The T, Banos O, Le BV, Bui DM, Yoon Y, Lee S - Sensors (Basel) (2015)

Bottom Line: As a key innovation, PAM captures not only the correlation among the activities, but also among the behaviors based on the arbitrary directed acyclic graph (DAG).The SVM classifier is then utilized on top to train and recognize the traffic activity and behavior.The proposed model shows more flexibility and greater expressive power than the commonly-used latent Dirichlet allocation (LDA) approach, leading to a higher recognition accuracy in the behavior classification.

View Article: PubMed Central - PubMed

Affiliation: Department of Computer Engineering, Kyung Hee University, Suwon 446-701, Korea. thienht@oslab.khu.ac.kr.

ABSTRACT
CCTV-based behavior recognition systems have gained considerable attention in recent years in the transportation surveillance domain for identifying unusual patterns, such as traffic jams, accidents, dangerous driving and other abnormal behaviors. In this paper, a novel approach for traffic behavior modeling is presented for video-based road surveillance. The proposed system combines the pachinko allocation model (PAM) and support vector machine (SVM) for a hierarchical representation and identification of traffic behavior. A background subtraction technique using Gaussian mixture models (GMMs) and an object tracking mechanism based on Kalman filters are utilized to firstly construct the object trajectories. Then, the sparse features comprising the locations and directions of the moving objects are modeled by PAMinto traffic topics, namely activities and behaviors. As a key innovation, PAM captures not only the correlation among the activities, but also among the behaviors based on the arbitrary directed acyclic graph (DAG). The SVM classifier is then utilized on top to train and recognize the traffic activity and behavior. The proposed model shows more flexibility and greater expressive power than the commonly-used latent Dirichlet allocation (LDA) approach, leading to a higher recognition accuracy in the behavior classification.

No MeSH data available.


Related in: MedlinePlus

Traffic activities discovered by PAM. (a–c) The vertical traffic behavior; (d–h) the horizontal traffic behavior.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4541867&req=5

f5-sensors-15-16040: Traffic activities discovered by PAM. (a–c) The vertical traffic behavior; (d–h) the horizontal traffic behavior.

Mentions: The experiments were performed on the QMUL (Queen Mary University of London) dataset [39], which includes a long-term video recorded at 25 fps for the frame rate and 360 × 288 for the frame resolution. Placed at an intersection, the video captured a busy traffic scenario involving a vehicle and pedestrian with dynamic movements. The video sequence was divided into short non-overlapping clips, each of which was 4 s. This duration is more convenient for observing when compared with too long a duration in Hospedales's work [40] (12 s) or two short a duration in Zhao's work [16] (2 s). The length of each clip was set to ensure that a behavior was not covered by others. A total of 750 clips comprised 320 vertical traffic flow clips; 430 horizontal traffic flow clips were tested with the manual activity and behavior labeling. Some activities cannot be fully categorized into horizontal or vertical traffic behavior, for example 40 frames may represent vertical traffic and 60 frames horizontal traffic. For example, a car can move in the vertical traffic from the top, and it will turn left or turn right at the intersection. Therefore, the authors categorized a given clip into either vertical or horizontal behavior based on the duration of the observed behaviors. If both behaviors are present during the whole clip, this is categorized into the most fluent behavior, i.e., with less changes or interruptions. In the vertical traffic, activities were discovered by PAM, as shown in Figure 5a–c. The horizontal traffic activities are presented in Figure 5d–h. Although PAM automatically discovered and modeled sparse words into super topics and subtopics, the number of topics had to be initially set. In this work, u = 2 for vertical and horizontal traffic behaviors; and v = 14 for traffic activities involving six vertical and eight horizontal activities. The description of the discovered activities outlined in Figure 5 is referenced in Table 2. In the PAM modeling, the Dirichlet distribution over behaviors and activities was produced with the parameter 0.01; the Gibbs sampling was processed with 1000 burn-in iterations. In the SVM-BTA classifier, the Gaussian kernel was used to set up for each node of binary classification. For each vertical and horizontal traffic dataset, the proposed method was evaluated using the 10-fold cross-validation. In order to analyze accuracy of the proposed method, Recall and Precision are used with the confusion matrix of each experiment. All of the experiments were performed on a desktop PC operating Windows 7 with a 2.67-GHz Intel Core i5 CPU and 4 GB of RAM. MATLAB R2013a was the software used for simulation.


Traffic Behavior Recognition Using the Pachinko Allocation Model.

Huynh-The T, Banos O, Le BV, Bui DM, Yoon Y, Lee S - Sensors (Basel) (2015)

Traffic activities discovered by PAM. (a–c) The vertical traffic behavior; (d–h) the horizontal traffic behavior.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4541867&req=5

f5-sensors-15-16040: Traffic activities discovered by PAM. (a–c) The vertical traffic behavior; (d–h) the horizontal traffic behavior.
Mentions: The experiments were performed on the QMUL (Queen Mary University of London) dataset [39], which includes a long-term video recorded at 25 fps for the frame rate and 360 × 288 for the frame resolution. Placed at an intersection, the video captured a busy traffic scenario involving a vehicle and pedestrian with dynamic movements. The video sequence was divided into short non-overlapping clips, each of which was 4 s. This duration is more convenient for observing when compared with too long a duration in Hospedales's work [40] (12 s) or two short a duration in Zhao's work [16] (2 s). The length of each clip was set to ensure that a behavior was not covered by others. A total of 750 clips comprised 320 vertical traffic flow clips; 430 horizontal traffic flow clips were tested with the manual activity and behavior labeling. Some activities cannot be fully categorized into horizontal or vertical traffic behavior, for example 40 frames may represent vertical traffic and 60 frames horizontal traffic. For example, a car can move in the vertical traffic from the top, and it will turn left or turn right at the intersection. Therefore, the authors categorized a given clip into either vertical or horizontal behavior based on the duration of the observed behaviors. If both behaviors are present during the whole clip, this is categorized into the most fluent behavior, i.e., with less changes or interruptions. In the vertical traffic, activities were discovered by PAM, as shown in Figure 5a–c. The horizontal traffic activities are presented in Figure 5d–h. Although PAM automatically discovered and modeled sparse words into super topics and subtopics, the number of topics had to be initially set. In this work, u = 2 for vertical and horizontal traffic behaviors; and v = 14 for traffic activities involving six vertical and eight horizontal activities. The description of the discovered activities outlined in Figure 5 is referenced in Table 2. In the PAM modeling, the Dirichlet distribution over behaviors and activities was produced with the parameter 0.01; the Gibbs sampling was processed with 1000 burn-in iterations. In the SVM-BTA classifier, the Gaussian kernel was used to set up for each node of binary classification. For each vertical and horizontal traffic dataset, the proposed method was evaluated using the 10-fold cross-validation. In order to analyze accuracy of the proposed method, Recall and Precision are used with the confusion matrix of each experiment. All of the experiments were performed on a desktop PC operating Windows 7 with a 2.67-GHz Intel Core i5 CPU and 4 GB of RAM. MATLAB R2013a was the software used for simulation.

Bottom Line: As a key innovation, PAM captures not only the correlation among the activities, but also among the behaviors based on the arbitrary directed acyclic graph (DAG).The SVM classifier is then utilized on top to train and recognize the traffic activity and behavior.The proposed model shows more flexibility and greater expressive power than the commonly-used latent Dirichlet allocation (LDA) approach, leading to a higher recognition accuracy in the behavior classification.

View Article: PubMed Central - PubMed

Affiliation: Department of Computer Engineering, Kyung Hee University, Suwon 446-701, Korea. thienht@oslab.khu.ac.kr.

ABSTRACT
CCTV-based behavior recognition systems have gained considerable attention in recent years in the transportation surveillance domain for identifying unusual patterns, such as traffic jams, accidents, dangerous driving and other abnormal behaviors. In this paper, a novel approach for traffic behavior modeling is presented for video-based road surveillance. The proposed system combines the pachinko allocation model (PAM) and support vector machine (SVM) for a hierarchical representation and identification of traffic behavior. A background subtraction technique using Gaussian mixture models (GMMs) and an object tracking mechanism based on Kalman filters are utilized to firstly construct the object trajectories. Then, the sparse features comprising the locations and directions of the moving objects are modeled by PAMinto traffic topics, namely activities and behaviors. As a key innovation, PAM captures not only the correlation among the activities, but also among the behaviors based on the arbitrary directed acyclic graph (DAG). The SVM classifier is then utilized on top to train and recognize the traffic activity and behavior. The proposed model shows more flexibility and greater expressive power than the commonly-used latent Dirichlet allocation (LDA) approach, leading to a higher recognition accuracy in the behavior classification.

No MeSH data available.


Related in: MedlinePlus