Crowd motion monitoring using tracklet-based commotion measure

Mousavi, Hossein; Nabi, Moin; Kiani, Hamed; Perina, Alessandro; Murino, Vittorio

doi:10.1109/icip.2015.7351223

Cited by 37 publications

(25 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Following the protocol suggested in [11], the predicted abnormal pixels are compared with the pixel-level ground truth. A test frame is a true positive Method AUC Optical-flow [14] 0.84 Social force (SFM) [14] 0.96 Sparse Reconstruction [3] 0.97 Commotion Measure [17] 0.98 TCP [26] 0.98 Adversarial Discriminator 0.99 Table 2. UMN dataset.…”

Section: Resultsmentioning

confidence: 99%

Training Adversarial Discriminators for Cross-Channel Abnormal Event Detection in Crowds

Ravanbakhsh

Sangineto

Nabi

et al. 2019

2019 IEEE Winter Conference on Applications of Computer Vision (WACV)

Self Cite

160

120

View full text Add to dashboard Cite

Abnormal crowd behaviour detection attracts a large interest due to its importance in video surveillance scenarios. However, the ambiguity and the lack of sufficient abnormal ground truth data makes end-to-end training of large deep networks hard in this domain. In this paper we propose to use Generative Adversarial Nets (GANs), which are trained to generate only the normal distribution of the data. During the adversarial GAN training, a discriminator (D) is used as a supervisor for the generator network (G) and vice versa. At testing time we use D to solve our discriminative task (abnormality detection), where D has been trained without the need of manually-annotated abnormal data. Moreover, in order to prevent G learn a trivial identity function, we use a cross-channel approach, forcing G to transform raw-pixel data in motion information and vice versa. The quantitative results on standard benchmarks show that our method outperforms previous state-of-the-art methods in both the frame-level and the pixel-level evaluation.

show abstract

Section: Resultsmentioning

confidence: 99%

Training Adversarial Discriminators for Cross-Channel Abnormal Event Detection in Crowds

Ravanbakhsh

Sangineto

Nabi

et al. 2019

2019 IEEE Winter Conference on Applications of Computer Vision (WACV)

Self Cite

160

120

View full text Add to dashboard Cite

show abstract

“…Toward this purpose, for each video block b i t a histogram h i t is computed to represent the distribution of prototypes in the video block. TCP Measure: Similarly to the commotion measure [24], to obtain the TCP measure for a given video block b i t , the irregularity of histogram h i t is computed. This is done by considering the fact that, if there is no difference in the appearance, then there is no change in descriptor features and consequently there is no change in the prototype representation.…”

Section: Temporal Cnn Pattern (Tcp)mentioning

confidence: 99%

“…We denote the TCP map for frame f t as c t = {c i t } I i=1 , where I is the number of patches in the frame. Method AUC Optical-Flow [21] 0.84 SFM [21] 0.96 Del Giorno et al [6] 0.910 Marsden et al [19] 0.929 Singh and Mohan [37] 0.952 Sparse Reconstruction [5] 0.976 Commotion [24] 0.988 Yu et al [44] 0.972 Cem et al [7] 0.964 TCP (proposed method) 0.988 Up-sampling TCP Maps: Since the frame will pass through several convolution and pooling layers in the network, the final TCP map is smaller than the original video frame. To localize the exact region there is a need to produce a map of the same size as the input frame.…”

Section: Temporal Cnn Pattern (Tcp)mentioning

confidence: 99%

“…Besides the insufficiency of data, the lack of a clear definition of abnormality (i.e., the context-dependent nature of the abnormality) induces subjectivity in annotations. Previous work highlighted the fact that the unsupervised measure-based methods may outperform supervised methods, due to the subjective nature of annotations as well as the small size of training data [42,38,41,24].…”

Section: Introductionmentioning

confidence: 99%

“…This leads to a more discriminative motion representation while maintaining the method complexity to a manageable level. Most related to our paper is the work by Mousavi et al [24], which introduced a similar measure to capture the commotion of a crowd motion for the task of abnormality detection. Instead of capturing the local irregularity of the low-level motion features (e.g., tracklets in [24]) or high-level detectors [25], we propose to represent the crowd motion exploiting the temporal variations of CNN features.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Plug-and-Play CNN for Crowd Motion Analysis: An Application in Abnormal Event Detection

Ravanbakhsh

Nabi

Mousavi

et al. 2018

2018 IEEE Winter Conference on Applications of Computer Vision (WACV)

Self Cite

159

117

View full text Add to dashboard Cite

Most of the crowd abnormal event detection methods rely on complex hand-crafted features to represent the crowd motion and appearance. Convolutional Neural Networks (CNN) have shown to be a powerful instrument with excellent representational capacities, which can leverage the need for hand-crafted features. In this paper, we show that keeping track of the changes in the CNN feature across time can be used to effectively detect local anomalies. Specifically, we propose to measure local abnormality by combining semantic information (inherited from existing CNN models) with low-level optical-flow. One of the advantages of this method is that it can be used without the fine-tuning phase. The proposed method is validated on challenging abnormality detection datasets and the results show the superiority of our approach compared with the state-of-theart methods.

show abstract