Analyzing motion patterns in crowded scenes via automatic tracklets clustering

Wang, Chongjing; Xu, Zeshui; Yi, Zou; Liu, Yuncai

doi:10.1109/cc.2013.6506940

Cited by 26 publications

(10 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The effectiveness of our proposed framework is compared with the relevant current approaches in the extraction of [10] 1812X1461 outdoor High-density Unimib [2] 480X360 Indoor Low-density Escalator [6] 480X480 Indoor Low-density Hajj [10] 2946X1656 Outdoor High-density Station [6] 957X719 Indoor Low-density dominant flows/global tracks and also in the identification of sources and sinks of the dominant flows. The extraction of accurate flows is vital to the understanding of crowd scene.…”

Section: Resultsmentioning

confidence: 99%

“…However, these approaches are not able to capture whole motion information because only few feature points are detected which are not sufficient for the extraction of dense trajectories. The most related work to our proposed work used KLT tracker to extract motion trajectories and used hierarchical clustering to obtain dominant flows [5], [6], [13]. However, the approaches are not able to capture the longterm motion required for understanding the behavior of the crowd.…”

Section: Literature Reviewmentioning

confidence: 99%

See 1 more Smart Citation

Extracting descriptive motion information from crowd scenes

Saqib

Khan²,

Sharma

et al. 2017

2017 International Conference on Image and Vision Computing New Zealand (IVCNZ)

View full text Add to dashboard Cite

Moreover, tracking is also not feasible in the high-density crowd. The tracker is lost easily and requires re-initialization at regular interval. The reason for suboptimal performance is that there is less number of pixels per person and inter-object occlusions, which make the detection and tracking unreliable solutions. Therefore in such scenarios, a holistic understanding of the scene is required. The study of crowd behavior includes the study of flows, dominant flows/common pathways, and identification of inflows/source and sinks/outflows. This study of behavior can be beneficial as a preprocessing step in the understanding of crowd behavior in extraordinary situations such as congestion.It is very critical for a computer vision system to capture accurate motion information for understanding the behavior of crowd. Thus the performance of the system is dependent upon how the motion is represented. The desired motion representation should generate long and most importantly reliable trajectories. These trajectories can then be used to describe the flow for the whole video. Traditionally, optical flow is used for computing the pixel-wise flow between the two frames [4], [12]. However, there are some inherent problems in the direct usage of optical flow for motion representation. Firstly, optical flow produces ambiguous results on the boundaries of conflicting flows and perform poorly in the case of very slow moving objects. Secondly, optical flow does not represent the long-range spatio-temporal motion representation required in many applications. Typically obtaining complete trajectories is a difficult task from crowd motion in the high-density crowd. Therefore to get complete trajectories a notion of tracklet is used to capture short-term motion. A tracklet is a fragment of trajectory obtained by the tracker for the object. As previously mentioned, detection of the object and their tracking does not perform well in a high-density crowd. In most of the previous studies, tracklets are extracted from feature points and subsequently tracked for the very short duration to generate tracklets. After tracklets generation, various approaches such as Markov Random Field (MRF) [22] are used to ensure spatio-temporal dependencies between the tracklets [22], [23]. The tracklets give a robust high-level representation of the crowd and are less like likely to drift as compared to complete trajectories [11].In this paper, we propose a framework for the study of Abstract-An important contribution that automated analysis tools can generate for management of pedestrians and crowd safety is the detection of conflicting l arge p edestrian fl ows: this kind of movement pattern, in fact, may lead to dangerous situations and potential threats to pedestrian's safety. For this reason, detecting dominant motion patterns and summarizing motion information from the scene are inevitable for crowd management. In this paper, we develop a framework that extracts motion information from the scene by generating point trajectories using particle a...

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Literature Reviewmentioning

confidence: 99%

Extracting descriptive motion information from crowd scenes

Saqib

Khan²,

Sharma

et al. 2017

2017 International Conference on Image and Vision Computing New Zealand (IVCNZ)

View full text Add to dashboard Cite

show abstract

“…Yang C et al [9] raised a new feature descriptor called multi-scale optical current histogram (MHOF) to retain continuous spatial information and motion information. Wang C et al [10] proposed to learn trajectory clustering of semantic regions, extract trajectories from dense feature points, and then use special models to enhance the spatio-temporal correlation between trajectories to detect pedestrian behavior patterns in crowded scenes. Moore et al [11] posed the opinion that people appear as particles in a fluid in certain aspects.…”

Section: Recent Workmentioning

confidence: 99%

3D Res-Inception Network Transfer Learning for Multiple Label Crowd Behavior Recognition

2019

KSII TIIS

View full text Add to dashboard Cite

The problem towards crowd behavior recognition in a serious clustered scene is extremely challenged on account of variable scales with non-uniformity. This paper aims to propose a crowed behavior classification framework based on a transferring hybrid network blending 3D res-net with inception-v3. First, the 3D res-inception network is presented so as to learn the augmented visual feature of UCF 101. Then the target dataset is applied to fine-tune the network parameters in an attempt to classify the behavior of densely crowded scenes. Finally, a transferred entropy function is used to calculate the probability of multiple labels in accordance with these features. Experimental results show that the proposed method could greatly improve the accuracy of crowd behavior recognition and enhance the accuracy of multiple label classification.

show abstract

“…They detected abnormal pedestrian flow characteristics with the help of location, speed and direction of a pedestrian. Wang Chongjing et al [5] proposed an approach for the analysis of motion pattern by clustering the tracklets using an unsupervised hierarchical clustering algorithm. Saxena et al [6] proposed a multiple-frame feature point detection and tracking based on KLT tracker.…”

Section: Related Workmentioning

confidence: 99%

“…Support vector machines (SVMs) will be designed for binary classification. Several binary classifiers are combined to construct a multiclass classifier [5]. The pairwise or one versus one approach is done since there is only few instances and hence time to train the classifier is much small.…”

Section: F Event Recognitionmentioning

confidence: 99%

An efficient method for crowd event recognition based on motion patterns

Sindhuja¹,

Srinivasagan²,

Kalaiselvi³

2014

2014 International Conference on Recent Trends in Information Technology

View full text Add to dashboard Cite

An automated visual monitoring process expands from low level analysis of object detection and tracking to the interpretation of their behaviors. Analyzing human crowd is an emerging trend in intelligent video surveillance for the purpose of detecting abnormalities. Tracking every human being in a crowd and analyzing their behavior is a challenging task due to occlusions. Hence, the crowd can be handled as a group entity instead of tracking the individual in the crowd. The behavior of the crowd can be distinguished with motion patterns due to prominent spatio-temporal characteristics. The proposed system involves a systematic approach to recognize the global events in human crowd through observing motion patterns such as flow, speed and direction. Initially as a preprocessing step, background subtraction is performed to extract the foreground blobs and optical flow is estimated to obtain the velocity and direction of motion. The human crowds are then clustered based on similar direction and proximity using Adjacency Matrix based Clustering (AMC). After clustering, the centroid and orientation of the cluster are extracted inorder to represent the behavior of crowd. Finally the multiclass Support Vector Machine (SVM) is trained to correctly recognize the behavior of crowd. Keywords-crowd behavior, optical flow, Adjacency Matrix based Clustering, Multiclass SVM.I.

show abstract

Analyzing motion patterns in crowded scenes via automatic tracklets clustering

Cited by 26 publications

References 0 publications

Extracting descriptive motion information from crowd scenes

Extracting descriptive motion information from crowd scenes

3D Res-Inception Network Transfer Learning for Multiple Label Crowd Behavior Recognition

An efficient method for crowd event recognition based on motion patterns

Contact Info

Product

Resources

About