Small-scale Pedestrian Detection Based on Somatic Topology Localization and Temporal Feature Aggregation

Song, Tao; Sun, Leiyu; Xie, Dong; Sun, Haiming; Pu, Shiliang

doi:10.48550/arxiv.1807.01438

Cited by 15 publications

(17 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Backbone Reasonable ACF [24] -51.68 Checkerboards [26] -39.67 Vanilla Faster R-CNN [16] VGG-16 20.00 Adapted Faster R-CNN [9] VGG-16 18.81 RPN+BF [29] VGG-16 23.26 SDS-RCNN [7] VGG-16 17.80 TFAN [15] ResNet-50 16.50 EGCL (ours) VGG-16 15.93 part [34], Adapted Faster R-CNN [9], OR-CNN [18], Adaptive-NMS [57], MGAN [11], HGPD [58], TLL [37], R 2 NMS [14], TLL+MRF [37], Replusion loss [17], ALFNet [19], NOH-NMS [60] and CrowdDet [59]. Note that these state-of-theart pedestrian detectors are trained on different backbones (VGG16 or ResNet-50) for feature learning head and different input scales (×1 or ×1.3).…”

Section: Methodsmentioning

confidence: 99%

“…MS-CNN [36] explores the different depth of feature maps from feature extractor to generate different sizes of proposals, which is followed by a detection network guided by context reasoning. Based on the assumption that different scales of pedestrians bodies can be modeled as 2D Gaussian kernel with various scale variance, TLL [37] designs a unified fully convolutional network to locate the somatic topological line of pedestrians with line annotation for detecting multi-scale pedestrians. Recently, ascribing the poor performance of detect small-scale pedestrians to the problem of inaccurate location, Cao et al [38] proposes a location bootstrap module for re-weighting the regression loss.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Pedestrian Detection by Exemplar-Guided Contrastive Learning

Lin¹,

Pei²,

Chen³

et al. 2021

Preprint

View full text Add to dashboard Cite

Section: Methodsmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Pedestrian Detection by Exemplar-Guided Contrastive Learning

Lin¹,

Pei²,

Chen³

et al. 2021

Preprint

View full text Add to dashboard Cite

“…Caltech dataset: Fig. 6(a) shows detection performance comparisons under Reasonable setup with state-of-theart methods including FasterRCNN+ATT [39], RPN+BF [35], AdaptFasterRCNN [37], PCN [30], F-DNN+SS [11], GDPL [23], F-DNN2+SS [12], TLL-TFA [27] and SDS-RCNN [3] on Caltech test set. It can be observed that the proposed SSA-CNN achieves the best detection performance with MR of 6.27% which is about 14.8% improvement over current top performing SDS-RCNN [3].…”

Section: Comparisons With State-of-the-art Methods On Benchmarksmentioning

confidence: 99%

“…PCN [30] utilizes two sub-networks which detects the pedestrians through body parts semantic information and context information respectively, and achieves MR 8.45% on Caltech dataset. In order to detect small-scale pedestrians, TLL+TFA [27] proposes a method that integrates somatic topological line localization (TLL) and temporal feature aggregation. TLL+TFA achieves MR 7.40% on Caltech dataset and obtains better detection performance for pedestrians that are far away from camera.…”

Section: Related Workmentioning

confidence: 99%

SSA-CNN: Semantic Self-Attention CNN for Pedestrian Detection

Zhou,

Wu,

Lam

2019

Preprint

View full text Add to dashboard Cite

Pedestrian detection plays an important role in many applications such as autonomous driving. We propose a method that explores semantic segmentation results as selfattention cues to significantly improve the pedestrian detection performance. Specifically, a multi-task network is designed to jointly learn semantic segmentation and pedestrian detection from image datasets with weak boxwise annotations. The semantic segmentation feature maps are concatenated with corresponding convolution features maps to provide more discriminative features for pedestrian detection and pedestrian classification. By jointly learning segmentation and detection, our proposed pedestrian selfattention mechanism can effectively identify pedestrian regions and suppress backgrounds. In addition, we propose to incorporate semantic attention information from multiscale layers into deep convolution neural network to boost pedestrian detection. Experiment results show that the proposed method achieves the best detection performance with MR of 6.27% on Caltech dataset and obtain competitive performance on CityPersons dataset while maintaining high computational efficiency.

show abstract

“…Nevertheless, we still compare the proposed approach with single segmentation or detection networks. We select several state-of-the-art segmentation networks (PSPNet, ICNet, BiSeNet) and detection networks (SSA-CNN [40], TLL [31], MFR-CNN [35]), and compare them with our network on Cityscape and Cityperson datasets. The results are given in the table 3.…”

Section: Backbonementioning

confidence: 99%

PIDNet: An Efficient Network for Dynamic Pedestrian Intrusion Detection

Sun

Chen

et al. 2020

Proceedings of the 28th ACM International Conference on Multimedia

View full text Add to dashboard Cite

Vision-based dynamic pedestrian intrusion detection (PID), judging whether pedestrians intrude an area-of-interest (AoI) by a moving camera, is an important task in mobile surveillance. The dynamically changing AoIs and a number of pedestrians in video frames increase the difficulty and computational complexity of determining whether pedestrians intrude the AoI, which makes previous algorithms incapable of this task. In this paper, we propose a novel and efficient multi-task deep neural network, PIDNet, to solve this problem. PIDNet is mainly designed by considering two factors: accurately segmenting the dynamically changing AoIs from a video frame captured by the moving camera and quickly detecting pedestrians from the generated AoI-contained areas. Three efficient network designs are proposed and incorporated into PIDNet to reduce the computational complexity: 1) a special PID task backbone for feature sharing, 2) a feature cropping module for feature cropping, and 3) a lighter detection branch network for feature compression. In addition, considering there are no public datasets and benchmarks in this field, we establish a benchmark dataset to evaluate the proposed network and give the corresponding evaluation metrics for the first time. Experimental results show that PIDNet can achieve 67.1% PID accuracy and 9.6 fps inference speed on the proposed dataset, which serves as a good baseline for the future vision-based dynamic PID study.

show abstract

Small-scale Pedestrian Detection Based on Somatic Topology Localization and Temporal Feature Aggregation

Cited by 15 publications

References 20 publications

Pedestrian Detection by Exemplar-Guided Contrastive Learning

Pedestrian Detection by Exemplar-Guided Contrastive Learning

SSA-CNN: Semantic Self-Attention CNN for Pedestrian Detection

PIDNet: An Efficient Network for Dynamic Pedestrian Intrusion Detection

Contact Info

Product

Resources

About