A Comparative Study of Transfer Learning Approaches for Video Anomaly Detection

Gutoski, Matheus; Ribeiro, Matilde; Hattori, Leandro Takeshi; Romero, Marcelo; Lazzaretti, André E.; Lopes, Heitor Silvério

doi:10.1142/s0218001421520030

Cited by 9 publications

(6 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Similarly, another eight hypotheses on the differences of this G 5 group are statistically significant, as their distance differences are greater than 5.8396 at 90% confidence limit. The performance of the method of AEcaUnet (Ours) is remarkably different than that of Gutoski et al [115], AEcUnet (Ours), Wu0S [98], and AE-Unet (Ours). Nevertheless, the performance of the method of Roy et al [91] is not remarkably different than that of Gutoski et al [115] and AEcUnet (Ours) at a confidence limit of 90%.…”

Section: Average Ranking Of Gcontrasting

confidence: 67%

See 1 more Smart Citation

Deep Crowd Anomaly Detection by Fusing Reconstruction and Prediction Networks

2023

View full text Add to dashboard Cite

Abnormal event detection is one of the most challenging tasks in computer vision. Many existing deep anomaly detection models are based on reconstruction errors, where the training phase is performed using only videos of normal events and the model is then capable to estimate frame-level scores for an unknown input. It is assumed that the reconstruction error gap between frames of normal and abnormal scores is high for abnormal events during the testing phase. Yet, this assumption may not always hold due to superior capacity and generalization of deep neural networks. In this paper, we design a generalized framework (rpNet) for proposing a series of deep models by fusing several options of a reconstruction network (rNet) and a prediction network (pNet) to detect anomaly in videos efficiently. In the rNet, either a convolutional autoencoder (ConvAE) or a skip connected ConvAE (AEc) can be used, whereas in the pNet, either a traditional U-Net, a non-local block U-Net, or an attention block U-Net (aUnet) can be applied. The fusion of both rNet and pNet increases the error gap. Our deep models have distinct degree of feature extraction capabilities. One of our models (AEcaUnet) consists of an AEc with our proposed aUnet has capability to confirm better error gap and to extract high quality of features needed for video anomaly detection. Experimental results on UCSD-Ped1, UCSD-Ped2, CUHK-Avenue, ShanghaiTech-Campus, and UMN datasets with rigorous statistical analysis show the effectiveness of our models.

show abstract

Section: Average Ranking Of Gcontrasting

confidence: 67%

“…From Figure 21, it is noticeable that the hypothesis on the difference of AEcaUnet (Ours) vs. Gutoski et al [115] is statistically significant. Similarly, another eight hypotheses on the differences of this G 5 group are statistically significant, as their distance differences are greater than 5.8396 at 90% confidence limit.…”

Section: Average Ranking Of Gmentioning

confidence: 80%

Deep Crowd Anomaly Detection by Fusing Reconstruction and Prediction Networks

2023

View full text Add to dashboard Cite

show abstract

“…The trained DNN is used for extracting features and, next, a new classifier is trained for the specific problem in hand. This approach was proved to be very useful in many computer vision applications [7], [20].…”

Section: B the Proposed Methodsmentioning

confidence: 99%

“…Deep Neural Networks (DNN) have achieved excellent results in solving several complex computer vision problems, for instance: object detection [5], object segmentation [6], anomaly detection in videos [7], people reidentification [8], gender and age recognition [9]. Following this idea, [10] proposed a deep learning model based on convolutional layers followed by LSTM layers to extract spatial and temporal features from videos.…”

Section: Introductionmentioning

confidence: 99%

Deep Learning for People Counting in Videos by Age and Gender

Inácio¹,

Ramos²,

Lopes³

2021

Anais Do 15. Congresso Brasileiro De Inteligência Computacional

View full text Add to dashboard Cite

Currently, many companies or even cities use surveillance cameras all the time, and due to the COVID-19 pandemic, many places have to limit the number of people in attendance. This paper proposes a method for people counting by gender and age in videos using deep learning techniques. The proposed method is based on a face detection and tracking approach combined with an alignment process to minimize the negative effect of the background information, considering occlusions and avoiding duplicate counting. Then, specialized Deep Neural Networks based on the EfficientNet architecture are employed for age and gender classification. Experimental results show that our method achieves satisfactory performance on people counting by age and gender, demonstrating the effectiveness of the present method.

show abstract

“…Following other works in the literature [23,24,27,47], we also used transfer learning to build a model for face mask classification. The base for our approach is the MobileNet-v2 model [48] previously trained with the ImageNet dataset.…”

Section: Face Mask Classificationmentioning

confidence: 99%

A Scalable Analytics Pipeline for COVID-19 Face Mask Surveillance

Kossoski¹,

Schaefer²,

Oliveira³

et al. 2022

Learn. Nonlin. Mod.

View full text Add to dashboard Cite

The COVID-19 coronavirus pandemic still causes a global health crisis. An effective protection method is using a face mask in public areas, according to the World Health Organization (WHO). Computer vision systems can be allies in monitoring public areas where the face mask is mandatory. However, face mask detection is challenging due to many factors, including diversity of people, facial features, head accessories, mask design, image position, and lighting changes. To tackle these issues, we present the following contributions: a new balanced face mask dataset named UTFPR-FMD1, consisting of 61,430 images splitted into “face” and “mask” classes; a transfer learning classification model for computer vision tasks, trained with our dataset; a new processing pipeline that allows face mask detection in video streams. Unlike available public datasets with imbalanced class distributions, the UTFPR-FMD1 contains images from different people, gender, and ages to minimize the training difficulty of deep learning models. We introduced a new measure to select valid images to perform inferences. Experimental results show the effectiveness of our model, outperforming the state-of-art methods for face mask detection tasks. Additionally, and different from other authors, we also present qualitative results. The system can detect heads with up to 60 degrees of rotation and process up to 10 FPS. In future work, we will deploy the current framework into production, perform tests in a near real-time environment, and extend it to process multiple video streams.

show abstract

A Comparative Study of Transfer Learning Approaches for Video Anomaly Detection

Cited by 9 publications

References 22 publications

Deep Crowd Anomaly Detection by Fusing Reconstruction and Prediction Networks

Deep Crowd Anomaly Detection by Fusing Reconstruction and Prediction Networks

Deep Learning for People Counting in Videos by Age and Gender

A Scalable Analytics Pipeline for COVID-19 Face Mask Surveillance

Contact Info

Product

Resources

About