Dynamic Region Division for Adaptive Learning Pedestrian Counting

Binhao Huang; Bin Sheng; Gaoqi He; Yubo Yuan; Zhenwei Ma

REVIEW 4 major objections 5 minor 25 references

Dynamic Region Division for Adaptive Learning Pedestrian Counting

T0 review · 4 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Dynamic region division that never cuts a head in half improves pedestrian counting on a five-scene subway benchmark.

desk verdict A hybrid detection/density counting pipeline with a genuinely dynamic mask, but the normalized density target in Eq. (4) means the reported MAE does not follow from the method as written. read the letter →

arxiv 1908.03978 v1 pith:SSZJZKN6 submitted 2019-08-12 cs.CV

Gaoqi He , Zhenwei Ma , Binhao Huang , Bin Sheng , Yubo Yuan This is my paper · ORCID

classification cs.CV

keywords pedestriancountingcrowddynamicregiondivisiondensitymapYOLOv3dilatedconvolutioninceptionmodulesurveillancevideo

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper argues that pedestrian counting in perspective-distorted surveillance video improves when the frame is not split by a fixed straight line but by a scene-adaptive boundary that keeps every detected head whole. It claims that a dynamic mask, built from YOLOv3 bounding boxes and an expected division height, lets a detector handle the near field and a density network handle the far field, and that fusing these counts yields lower average counting error than the straight-line double-region method and than two deep baselines on a five-scene subway dataset. A reader should care because the failure mode addressed—cutting one head across a region boundary and fitting large heads with Gaussian kernels—is a real obstacle for crowd counting in high-perspective scenes.

What carries the argument

The carrying mechanism is the dynamic division mask $M$: a binary image, produced by Algorithm 1 from YOLOv3 bounding boxes and expectation height $H$, that assigns each pixel to distant (1) or nearby (0) region while preserving head completeness. The mask also supplies the perspective map via linear regression on detected boxes, setting $\sigma = 0.15 M(p)$ for Gaussian head kernels in the distant region. The other named component, IDCNN, concatenates three dilated convolution branches (rates 1, 2, 3) in inception layers so the network learns which dilation rate to use instead of requiring the rate to be chosen a priori.

What would settle it

Re-run the pipeline on the same five subway scenes after disabling the detector's bounding boxes and drawing the boundary as a straight line at the same expectation height; if the average MAE does not rise materially above 2.30, the dynamic division itself is not what is carrying the improvement. Alternatively, in a scene where YOLOv3 is known to miss a cluster of nearby pedestrians, the mask will fail to include those heads, and counting that frame will show whether the missed detections propagate into the distant-region density estimate.

Watch

Extended reading notes

Core claim

The paper's central discovery is that the region boundary itself is part of the counting problem: drawing it as a straight line through a perspective-distorted scene splits heads and corrupts both the detection and density estimates. Its dynamic region division uses the detected bounding boxes to compute an expectation height $H$, then generates a mask that follows the tops and sides of those boxes so every head remains either entirely in the nearby region or entirely in the distant region. With this mask, the nearby region is counted by a pedestrian detector and the distant region by an inception dilated convolutional network (IDCNN), and the fused result reports average MAE 2.30 over the five subway scenes, the lowest among the compared methods.

Load-bearing premise

The load-bearing premise is that the detector finds every nearby pedestrian's head and places it correctly, because a missed or misplaced detection makes the region mask, the head-completeness guarantee, and the Gaussian kernel widths for distant heads all wrong.

Editorial extensions

If this is right

On the subway station dataset, the dynamic-division pipeline reports average MAE 2.30, lower than the straight-line double-region method (2.75), MCNN (2.69), and CSRNet (2.82), with the best per-scene result on three of the five scenes.
Fusing detection for the nearby region and density estimation for the distant region avoids using Gaussian kernels on large heads and avoids detector failures in dense distant crowds, so the method extends to scenes where neither approach alone is reliable.
The dynamic mask can be recomputed per frame, so the region division adapts to pedestrian layout rather than being fixed by camera geometry.
IDCNN removes the need to tune dilation rate by letting the network combine multiple rates, which can be applied to other counting or segmentation tasks that use dilated convolutions.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

If the claim holds, the division boundary, not just the counting network, is a major source of error in crowd counting; future systems could learn the boundary jointly with the counting objective rather than deriving it from detections.
A testable extension would replace YOLOv3 boxes with an oracle detector to isolate how much of the MAE gain comes from the mask itself versus the IDCNN, since both depend on the same detections.
The same mask idea could generalize to other regression-plus-detection hybrids, such as vehicle counting or cell counting in images with large scale variation.
The paper's per-scene failure in occlusion-heavy scenes suggests that dynamic division could be extended to incorporate occlusion cues, not just head completeness, when choosing the boundary.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague,

I read the He et al. paper on dynamic region division for pedestrian counting. The core idea is a modest but real step: use YOLOv3 detections to shape the boundary between a nearby region (counted by detection) and a distant region (counted by a density network), so heads don't get bisected by a straight line. The inception-style dilated convolution with parallel dilation rates is a reasonable workaround to the dilation-rate selection problem. The reported average MAE (2.30) is better than the three baselines on the five-scene subway dataset, and the comparison includes external methods (MCNN, CSRNet), not just their own prior work. So far so good.

The problem is in the details. Equation (4) defines the ground-truth density map as a sum of normalized Gaussians weighted by 1/||Pi||. Each Gaussian integrates to 1, so the integral of Di(p) is exactly 1 for every frame, no matter how many pedestrians are in the scene. The paper never explains how the pedestrian count is recovered from that map. The MAE in Eq. (5) compares counts, and their figures show counts like 117 vs 116.1. If the prediction is the sum of the density map, it would saturate near 1 and the MAE would be about 28, not 2.30. If some rescaling is applied, it is unstated and would need the true count at test time. That is a load-bearing inconsistency, not a typo: every training and test result depends on the definition of the target density.

The YOLOv3 dependency is a softer concern. The mask and the perspective map M(p) are both fitted from detector boxes, so missed or mislocated detections corrupt the region boundary and the Gaussian sigmas. That is an engineering trade-off, and a proper ablation would show how sensitive the result is. There are no error bars, no significance tests, and no code or annotations released. The method also loses to MCNN in two of the five scenes, so the average gain is fragile.

The citation pattern is fine; self-citing the prior straight-line method as baseline is appropriate.

Who should read this? Someone building hybrid detection/counting systems for fixed cameras would find the dynamic mask idea worth considering. But the unexplained normalization means the central claim is not supported as written. I would want a revision that derives the count extraction step and provides per-scene variance before trusting the 2.30 number.

Recommendation: send it to peer review — the idea is worth exploring and the flaw is fixable — but reviewers should push hard on the density normalization before accepting any of the quantitative claims.

Referee Report

4 major / 5 minor

Summary. The paper proposes a hybrid pedestrian-counting pipeline for subway surveillance frames. A dynamic boundary, computed from YOLOv3 detections and an expected head-height line H, splits each frame into a nearby region counted by YOLOv3 and a distant region counted by a newly designed inception dilated CNN (IDCNN); the two counts are fused to produce a per-frame total. Experiments on a five-scene subway dataset report an average MAE of 2.30, below the compared baselines He et al. [7], MCNN, and CSRNet. The central claim is that the dynamic region division plus IDCNN yields a lower average MAE than straight-line double-region division and the external baselines.

Significance. The underlying design idea is reasonable and of practical interest: instead of fitting a single density estimator to a scene with highly variable head scale, the paper uses detection where pedestrians are large and density regression where they are small, and it attempts to make the boundary respect head integrity. The multi-dilation inception module that lets the network learn the effective dilation rate is a simple and sensible contribution. The comparison includes external baselines (MCNN, CSRNet) in addition to the authors' prior method [7], which is a useful feature of the experiments. If the empirical results were reproducible, the method would be a credible baseline for scene-adaptive counting. However, the evaluation pipeline is not internally consistent as written, the mask-generation pseudocode is not reproducible, and no ablation isolates the claimed sources of improvement.

major comments (4)

[§4.1, Eq. (4)] As written, Eq. (4) defines D_i(p) = Σ_{P∈P_i} (1/‖P_i‖) N(p;P,σ), where N is a normalized Gaussian. Because each term integrates to unity, the spatial integral of D_i is exactly 1 for every frame, independent of the pedestrian count. The paper never explains how the scalar count z_i in Eq. (5) is recovered from this normalized map. If z_i is the integral or sum of the network output, the estimate is constrained near 1 per frame while the dataset average is 28.78, making the reported MAE of 2.30 impossible. If an additional rescaling is applied, it is not described and would require the test-time count to be known, leaking annotation information. This is a load-bearing inconsistency in the training/evaluation definition, and Table 1 cannot be interpreted until it is resolved.
[§3.2, Algorithm 1] The pseudocode for the dynamic mask contains apparent mistakes that prevent reproduction of the division. In the non-overlap branch, line 14 re-assigns the same submatrix as line 11, and line 15 unconditionally marks columns from br x_{i-1} to Iwidth as distant, which includes the current bounding box and all boxes to its right. This contradicts the stated head-completeness guarantee and the illustration in Figure 3. The overlap branch is also unexplained: it uses tf y_{i-1} rather than H for the upper boundary. Please correct the pseudocode and, ideally, release the mask-generation code so the proposed division rule can be checked.
[§5.3, Table 1] The paper attributes the performance gain to dynamic region division and the IDCNN, but the comparison is not controlled. There is no ablation that replaces the dynamic division with a straight-line division while keeping the same YOLOv3/IDCNN subsystems, and no ablation that removes the IDCNN or the inception module. The reported average improvement over MCNN is 2.30 versus 2.69, yet in Jing'an Temple and South Railway Station the proposed method is worse than MCNN (1.59 versus 1.44 and 2.81 versus 2.35, respectively), and no variance estimates or significance tests are provided. As presented, the specific contribution of the dynamic division is not established.
[§4.1 and §3.2] The Gaussian width σ = 0.15 M(p) is obtained by linear regression from YOLOv3 bounding boxes, and the dynamic mask itself is built from those boxes. If YOLOv3 misses or mislocates pedestrians, the distant-region ground truth, the region boundary, and the head-completeness guarantee are all affected. The paper does not report YOLOv3 recall or precision on the nearby region, nor does it analyze the sensitivity of the final counts to detector errors. Because the training target and the region division both depend on the same detector, this is a load-bearing dependency that needs explicit validation.

minor comments (5)

[§5.2, Eq. (5)] The MAE formula should be written with an explicit parenthesis: MAE = (1/N) Σ_{i=1}^N |z_i − z_i^gt|; as printed, the markup is ambiguous.
[§4.1] The linear-regression step for the perspective map M(p) is described in one sentence; please specify the input features, the number of training boxes, and how M(p) is extrapolated to pixels without detected boxes.
[Abstract and Introduction] The word 'Ulteriorly' appears to be a word-choice error; 'subsequently' or 'furthermore' would be clearer.
[Figure 5 caption] The phrase 'The third row column' is garbled; it should say 'the third column'.
[§5.3] The discussion of occlusion in the Jing'an Temple and South Railway Station scenes is qualitative; provide quantitative occlusion statistics or at least example frames if this explanation is retained.

Circularity Check

1 steps flagged · score 6.0 of 10

Eq. (4) normalizes the count out of the density target, so the reported MAE reduces to a rescaling by the ground-truth count; otherwise the result is an empirical comparison with non-circular external baselines.

self definitional [Section 4.1, Eq. (4); Section 5.2, Eq. (5); Section 5.3 reported counts (GT 117.0, ET 116.1)]
"The ground truth density map is created as: Di(p) = ∑_{P∈Pi} 1/‖Pi‖ N(p;P,σ) (4) ... ‖Pi‖ is the number of annotated heads and the whole distribution is normalized by its reciprocal. ... MAE = 1/N ∑ |zi−zgt_i| (5) where ... zi is the estimated number of pedestrians in the i-th frame."

Each N is a normalized 2D Gaussian, so each head contributes integral 1; summing over ‖Pi‖ heads and multiplying by 1/‖Pi‖ makes the ground-truth density map integrate to exactly 1 for every frame, independent of the pedestrian count. The count is divided out of the target. Yet Eq. (5) evaluates a count error and Section 5.3 reports estimates such as GT 117.0 vs ET 116.1. No count-recovery step is given between the network's density output and zi. If zi is obtained by integrating the estimated density map, the predicted integral is ~1, not ~117.

full rationale

The paper's headline claim is an empirical MAE comparison on the subway dataset. The proposed dynamic-region division and IDCNN are compared with the authors' prior straight-line method [7], MCNN [9], and CSRNet [11]. Using [7] as a baseline and dataset source is a self-citation, but it is not load-bearing for the claimed improvement because independent external baselines are included and the dynamic-division algorithm is a new contribution. The perspective map M(p) fitted from YOLOv3 boxes and used to set σ=0.15M(p) is a dependency on the detector, not a circular step, since it does not by itself force the final count. There is, however, one genuine construction-level circular problem: Eq. (4) normalizes the ground-truth density map by 1/‖Pi‖, making the target integrate to 1 per frame. The count is then not recoverable from the density output without multiplying by the annotation count, which is the quantity the paper claims to predict. Unless an unstated rescaling is applied, the reported ET values (e.g., 116.1 vs GT 117.0) cannot be produced from the trained density map; with the rescaling, the MAE is forced by the known count. This makes the evaluation chain partially circular as written, though the rest of the framework (dynamic mask, IDCNN architecture, detection-count fusion) has independent content.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

All three support structures are present: alpha and the sigma scale are hand-set constants, the perspective map is fit from detector outputs rather than calibrated independently, and the key division line H is defined from YOLOv3 detections. The distant-region density ground truth depends on these fitted quantities. No invented physical entities are introduced.

free parameters (3)

alpha = 0.3
Hand-set head-to-body proportion used in Section 3.2, step (4) to determine which bounding boxes contain heads that would be cut by a straight division line.
sigma scale coefficient = 0.15
Section 4.1 sets sigma = 0.15 M(p); the 0.15 multiplier is chosen without sensitivity analysis or per-scene calibration.
Perspective map M(p) regression coefficients = not reported
Section 4.1 obtains M(p) from linear regression on YOLOv3 detected bounding boxes in the nearby region; these coefficients determine the Gaussian widths in the distant-region ground truth.

assumptions (5)

domain assumption A 2D Gaussian kernel with sigma proportional to a perspective map can model a distant pedestrian head.
Invoked in Section 4.1, Eq. (4), where the distant-region ground truth is generated as normalized Gaussians at annotated head centers with sigma = 0.15 M(p).
domain assumption YOLOv3 detector provides sufficiently accurate bounding boxes for nearby pedestrians to define the dynamic division mask and the perspective map.
Section 3.2 uses detected boxes as input to Algorithm 1; Section 4.1 uses detected boxes in the nearby region to fit M(p) by linear regression.
ad hoc to paper The expectation height H, computed as the probability-weighted average of pedestrian center heights, is a valid division line for the scene.
Section 3.2, Eqs. (2)-(3) define H from detected pedestrian positions; no independent justification is given for why this statistic separates nearby and distant regions.
ad hoc to paper alpha = 0.3 approximates the head-to-body proportion for all pedestrians in all five scenes.
Section 3.2, step (4) uses (1-alpha)*tf_y + alpha*br_y to find heads that would be cut; alpha is set to 0.3 without per-scene calibration or sensitivity analysis.
domain assumption Splitting the scene into a nearby detection region and a distant density region improves counting accuracy.
Borrowed from He et al. [7]; the paper does not re-test this premise against a single-region baseline.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Dynamic Region Division for Adaptive Learning Pedestrian Counting." pith.science (2026). https://pith.science/paper/SSZJZKN6

@misc{pith2026190803978,
  author       = {Pith},
  title        = {Pith review of: Dynamic Region Division for Adaptive Learning Pedestrian Counting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SSZJZKN6}},
  note         = {Machine review of arXiv:1908.03978}
}

read the original abstract

Accurate pedestrian counting algorithm is critical to eliminate insecurity in the congested public scenes. However, counting pedestrians in crowded scenes often suffer from severe perspective distortion. In this paper, basing on the straight-line double region pedestrian counting method, we propose a dynamic region division algorithm to keep the completeness of counting objects. Utilizing the object bounding boxes obtained by YoloV3 and expectation division line of the scene, the boundary for nearby region and distant one is generated under the premise of retaining whole head. Ulteriorly, appropriate learning models are applied to count pedestrians in each obtained region. In the distant region, a novel inception dilated convolutional neural network is proposed to solve the problem of choosing dilation rate. In the nearby region, YoloV3 is used for detecting the pedestrian in multi-scale. Accordingly, the total number of pedestrians in each frame is obtained by fusing the result in nearby and distant regions. A typical subway pedestrian video dataset is chosen to conduct experiment in this paper. The result demonstrate that proposed algorithm is superior to existing machine learning based methods in general performance.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

25 extracted references · 23 canonical work pages

[7]

The novel dynamic region division can meet the challenge of perspective distortion and avoid to cut the head into two parts

CONCLUSION In this paper, we propose an dynamic region learning algo- rithm for pedestrian counting in subway surveillance videos. The novel dynamic region division can meet the challenge of perspective distortion and avoid to cut the head into two parts. In the nearby region, we retrain YOLOV3 detector to substi- tute for using inaccurate gaussian kernel...

work page
[1]

Dynamic Region Division for Adaptive Learning Pedestrian Counting

INTRODUCTION Crowd gathering lead to huge loss of people’s life, property and cause terrible social inﬂuence [1]. Subway station, rail- way station and scenic area are typical crowd gathering scene. In Shanghai, China, the daily average passenger volume in subway station comes to 10,330,000 in 2018 and increases up more than 11% from a year earlier. There...

work page Pith review arXiv 2018
[2]

2) Appropriate learn- ing models are applied to count pedestrians in each obtained region

Utilizing the object bounding boxes obtained by YoloV3 and expectation division line of the scene, the boundary for nearby region and distant one is generated under the premise of keeping the completeness of heads. 2) Appropriate learn- ing models are applied to count pedestrians in each obtained region. In distant region, a novel inception dilated convo-...

work page
[3]

[2] and used 2D gaussian kernel to model one pedestrian

RELA TED WORK Counting by density map Density map was ﬁrst introduced into pedestrian counting ﬁeld by Lempitsky et al. [2] and used 2D gaussian kernel to model one pedestrian. Then Fiaschi et al. [6] used random forest to regress the object density and im- proved training efﬁciency. With the powerful ability of deep learning, Zhang et al. [8] ﬁrst explor...

work page
[4]

Overview To overcome the error of gaussian kernel simulation and misidentiﬁcation of clutter background caused by perspective distortion, a novel algorithm framwork is proposed

DYNAMIC REGION DIVISION ALGORITHM 3.1. Overview To overcome the error of gaussian kernel simulation and misidentiﬁcation of clutter background caused by perspective distortion, a novel algorithm framwork is proposed. Figure 2 clearly shows the ﬂow chart of algorithm. Since we ﬁnd that gaussian kernels are not suitable for simulating large heads, basing on...

work page
[5]

Counting model for distant region Li et al

COUNTING MODEL 4.1. Counting model for distant region Li et al. [11] proposed a CSRNet for crowd counting which introduced dilated convolution to improve traditional convo- lutional neural network. Dilated convolution enlarges the re- ceptive ﬁeld without increasing the number of parameters or the amount of computation. However, it is hard to choose the d...

work page
[6]

Experiment dataset We evaluate the proposed algorithm through extensive exper- iments on the publicly available Subway station pedestrian dataset [7]

EXPERIMENT 5.1. Experiment dataset We evaluate the proposed algorithm through extensive exper- iments on the publicly available Subway station pedestrian dataset [7]. The dataset covers ﬁve typical subway station scenes in Shanghai. These scenes mainly locates at the trans- fer corridors, which have severe perspective distortion and large variance of head...

work page
[8]

Research on the impact of crowd ﬂow on crowd risk in large gathering spots,

Zimei Liu, Yun Chen, and Kefan Xie, “Research on the impact of crowd ﬂow on crowd risk in large gathering spots,” in Industrial Informatics-Computing Technol- ogy, Intelligent Technology, Industrial Information In- tegration (ICIICII), 2016 International Conference on . IEEE, 2016, pp. 368–371

work page 2016

Show all 25 references

[9]

Learning to count objects in images,

Victor Lempitsky and Andrew Zisserman, “Learning to count objects in images,” in Advances in neural infor- mation processing systems, 2010, pp. 1324–1332

2010
[10]

Learning a perspective- embedded deconvolution network for crowd counting,

Muming Zhao, Jian Zhang, Fatih Porikli, Chongyang Zhang, and Wenjun Zhang, “Learning a perspective- embedded deconvolution network for crowd counting,” in Multimedia and Expo (ICME), 2017 IEEE Interna- tional Conference on. IEEE, 2017, pp. 403–408

2017
[11]

Histograms of ori- ented gradients for human detection,

Navneet Dalal and Bill Triggs, “Histograms of ori- ented gradients for human detection,” in Computer Vi- sion and Pattern Recognition, 2005. CVPR 2005. IEEE Computer Society Conference on . IEEE, 2005, vol. 1, pp. 886–893

2005
[12]

You only look once: Uniﬁed, real-time object detection,

Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi, “You only look once: Uniﬁed, real-time object detection,” in Proceedings of the IEEE confer- ence on computer vision and pattern recognition, 2016, pp. 779–788

2016
[13]

Learning to count with regression forest and structured labels,

Luca Fiaschi, Ullrich K ¨othe, Rahul Nair, and Fred A Hamprecht, “Learning to count with regression forest and structured labels,” in Pattern Recognition (ICPR), 2012 21st International Conference on. IEEE, 2012, pp. 2685–2688

2012
[14]

A double-region learning algorithm for counting the number of pedestrians in subway surveil- lance videos,

Gaoqi He, Qi Chen, Dongxu Jiang, Xingjian Lu, and Yubo Yuan, “A double-region learning algorithm for counting the number of pedestrians in subway surveil- lance videos,” Engineering Applications of Artiﬁcial In- telligence, vol. 64, pp. 302–314, 2017

2017
[15]

Cross-scene crowd counting via deep convolutional neural networks,

Cong Zhang, Hongsheng Li, Xiaogang Wang, and Xi- aokang Yang, “Cross-scene crowd counting via deep convolutional neural networks,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2015, pp. 833–841

2015
[16]

Single-image crowd counting via multi-column convolutional neural network,

Yingying Zhang, Desen Zhou, Siqin Chen, Shenghua Gao, and Yi Ma, “Single-image crowd counting via multi-column convolutional neural network,” in Pro- ceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 589–597

2016
[17]

Body structure aware deep crowd counting,

Siyu Huang, Xi Li, Zhongfei Zhang, Fei Wu, Shenghua Gao, Rongrong Ji, and Junwei Han, “Body structure aware deep crowd counting,”IEEE Transactions on Im- age Processing, vol. 27, no. 3, pp. 1049–1059, 2018

2018
[18]

Csr- net: Dilated convolutional neural networks for under- standing the highly congested scenes,

Yuhong Li, Xiaofan Zhang, and Deming Chen, “Csr- net: Dilated convolutional neural networks for under- standing the highly congested scenes,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 1091–1100

2018
[19]

Divide and grow: Captur- ing huge diversity in crowd images with incrementally growing cnn,

Deepak Babu Sam, Neeraj N Sajjan, R Venkatesh Babu, and Mukundhan Srinivasan, “Divide and grow: Captur- ing huge diversity in crowd images with incrementally growing cnn,” in Conference on Computer Vision and Pattern Recognition, 2018

2018
[20]

Cnn-based cascaded multi-task learning of high-level prior and den- sity estimation for crowd counting,

Vishwanath A Sindagi and Vishal M Patel, “Cnn-based cascaded multi-task learning of high-level prior and den- sity estimation for crowd counting,” in Advanced Video and Signal Based Surveillance (AVSS), 2017 14th IEEE International Conference on. IEEE, 2017, pp. 1–6

2017
[21]

Rich feature hierarchies for accurate object detection and semantic segmentation,

Ross Girshick, Jeff Donahue, Trevor Darrell, and Jiten- dra Malik, “Rich feature hierarchies for accurate object detection and semantic segmentation,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2014, pp. 580–587

2014
[22]

Ssd: Single shot multibox detector,

Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, and Alexander C Berg, “Ssd: Single shot multibox detector,” inEuropean conference on computer vision. Springer, 2016, pp. 21– 37

2016
[23]

Yolov3: An incremental improvement,

Joseph Redmon and Ali Farhadi, “Yolov3: An incremental improvement,” arXiv preprint arXiv:1804.02767, 2018

2018 arXiv
[24]

Going deeper with convolutions,

Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Ser- manet, Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, and Andrew Rabinovich, “Going deeper with convolutions,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2015, pp. 1–9

2015
[25]

Microsoft coco: Common objects in context,

Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll ´ar, and C Lawrence Zitnick, “Microsoft coco: Common objects in context,” inEuropean conference on computer vision. Springer, 2014, pp. 740–755

2014

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [7]

The novel dynamic region division can meet the challenge of perspective distortion and avoid to cut the head into two parts

CONCLUSION In this paper, we propose an dynamic region learning algo- rithm for pedestrian counting in subway surveillance videos. The novel dynamic region division can meet the challenge of perspective distortion and avoid to cut the head into two parts. In the nearby region, we retrain YOLOV3 detector to substi- tute for using inaccurate gaussian kernel...

work page

[2] [1]

Dynamic Region Division for Adaptive Learning Pedestrian Counting

INTRODUCTION Crowd gathering lead to huge loss of people’s life, property and cause terrible social inﬂuence [1]. Subway station, rail- way station and scenic area are typical crowd gathering scene. In Shanghai, China, the daily average passenger volume in subway station comes to 10,330,000 in 2018 and increases up more than 11% from a year earlier. There...

work page Pith review arXiv 2018

[3] [2]

2) Appropriate learn- ing models are applied to count pedestrians in each obtained region

Utilizing the object bounding boxes obtained by YoloV3 and expectation division line of the scene, the boundary for nearby region and distant one is generated under the premise of keeping the completeness of heads. 2) Appropriate learn- ing models are applied to count pedestrians in each obtained region. In distant region, a novel inception dilated convo-...

work page

[4] [3]

[2] and used 2D gaussian kernel to model one pedestrian

RELA TED WORK Counting by density map Density map was ﬁrst introduced into pedestrian counting ﬁeld by Lempitsky et al. [2] and used 2D gaussian kernel to model one pedestrian. Then Fiaschi et al. [6] used random forest to regress the object density and im- proved training efﬁciency. With the powerful ability of deep learning, Zhang et al. [8] ﬁrst explor...

work page

[5] [4]

Overview To overcome the error of gaussian kernel simulation and misidentiﬁcation of clutter background caused by perspective distortion, a novel algorithm framwork is proposed

DYNAMIC REGION DIVISION ALGORITHM 3.1. Overview To overcome the error of gaussian kernel simulation and misidentiﬁcation of clutter background caused by perspective distortion, a novel algorithm framwork is proposed. Figure 2 clearly shows the ﬂow chart of algorithm. Since we ﬁnd that gaussian kernels are not suitable for simulating large heads, basing on...

work page

[6] [5]

Counting model for distant region Li et al

COUNTING MODEL 4.1. Counting model for distant region Li et al. [11] proposed a CSRNet for crowd counting which introduced dilated convolution to improve traditional convo- lutional neural network. Dilated convolution enlarges the re- ceptive ﬁeld without increasing the number of parameters or the amount of computation. However, it is hard to choose the d...

work page

[7] [6]

Experiment dataset We evaluate the proposed algorithm through extensive exper- iments on the publicly available Subway station pedestrian dataset [7]

EXPERIMENT 5.1. Experiment dataset We evaluate the proposed algorithm through extensive exper- iments on the publicly available Subway station pedestrian dataset [7]. The dataset covers ﬁve typical subway station scenes in Shanghai. These scenes mainly locates at the trans- fer corridors, which have severe perspective distortion and large variance of head...

work page

[8] [8]

Research on the impact of crowd ﬂow on crowd risk in large gathering spots,

Zimei Liu, Yun Chen, and Kefan Xie, “Research on the impact of crowd ﬂow on crowd risk in large gathering spots,” in Industrial Informatics-Computing Technol- ogy, Intelligent Technology, Industrial Information In- tegration (ICIICII), 2016 International Conference on . IEEE, 2016, pp. 368–371

work page 2016