Deep Crowd Anomaly Detection by Fusing Reconstruction and Prediction Networks

Round 1

Reviewer 1 Report

Detection of abnormal events in automated video surveillance systems is one of the most challenging, overriding, and time-sensitive tasks. Deep learning based algorithms have been dominating the literature as the deep learning solutions for crowd events detection have outperformed the conventional machine learning solutions. Motion and appearance features are widely used in video anomaly detection algorithms. In deep learning based video anomaly detection algorithms, a common technique is to build reconstruction model considering motion and/or appearance features. A common assumption is that the reconstruction error of the frame of normal event is small but that of the frame of abnormal event is large. The authors analyze the specialized literature, generating lines of research that can bring a substantial scientific contribution to the researched field. This research is supported by a series of important bibliographic references.The authors propose six deep models from a generalized architecture by merging several others prediction and reconstruction networks to effectively detect anomalies in video. mesh fusion guaranteed some degree of reconstruction error increase gap. Experiments on five benchmark datasets demonstrated the potential of our models, and the detailed discussion verified their effectiveness to detect abnormal video events. Some of the models have shown promising results in terms of their ability to extract good quality a characteristics.

 

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The overall work is interesting. However, the concise presentation and overall discussion are not good enough. Further improvement is required as listed below.

1. The overall framework in figure 1 should be discussed and compared if reconstruction is required for the abnormal detection and decision making. It can also be compared with feature extraction, selection and fusion approach e.g. M Roopak, etc., Multi‐objective‐based feature selection for DDoS attack detection in IoT networks, IET Networks, 2020;

2. Mind Figures' quality and font e.g. normal/abnormal output should use diamond shape in the block diagram of Figure 1; Figures 3, 5, 6, 7 should have better font size and discussion;

3. Mind concise presentation and critical discussion with quantitative analysis.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper is overall well-written and has some merits.

I only have some concerns about the references. It seems that some important and relevant references are missing.

Frequency Feature Pyramid Network With Global-Local Consistency Loss for Crowd-and-Vehicle Counting in Congested Scenes, IEEE Transactions on Intelligent Transportation Systems

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

In my opinion the work is very well presented, has a good review, the application has high merit. Considered that it is in the condition of being accepted without modification.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf