PedSurvey

From Handcrafted to Deep Features for Pedestrian Detection: A Survey

This project provides a paper list about pedestrian detection following the taxonomy in "From Handcrafted to Deep Features for Pedestrian Detection: A Survey (IEEE TPAMI 2022)".

• Single-spectral pedestrian detection and multispectral pedestrian detection are both summarized.
• The performance of some methods on different datasets are shown in Leaderboard.
• We release a new large-scae pedestrian detection dataset TJU-DHD-Pedestrian: arXiv2021, TIP2021, website, Learderboard
• If you find a new paper or an error, please feel free to contact us.

News

**PD**: Pedestrian Detection; **MPD**: Multispectral Pedestrian Detection; **MVD**: Multi-View Pedestrian Detection; **Others**: Pedestrian Detection with Special Devices

• [Aug. 18, 2023]: PD: OTP-NMS(TIP2023), Seq2SeqNMS(TIP2023), MsSE-SR(TITS2023), CFRLA-Net(TCSVT2023); MPD: MCHE-CF(TMM2023); Others: PEDRo(CVPRW2023), NIRPed(TITS2023)
• [June 9, 2023]: PD: OPL(CVPR2023), LSFM(CVPR2023), VLPD(CVPR2023);
• [April 16, 2023]: PD: CTD(TMM2023), DDAD(TITS2023), MMPD-MDCNN(IF2023), DINF(arXiv2023); Others: PiFeNet(RAL2023)
• [April 16, 2023]: We add Multi-View Pedestrian Detection (MVD) below and also present some works here. MVD: MVAug(WACV2023), KSMVD(PR2022), 3DROM(ECCV2022), SHOT(ICCV2021), MVDeTr(ACMMM2021), MVDet(ECCV2020)
• [Dec. 31, 2022]: PD: SMFE(NeurIPS2022), MB-CSP(TITS2022), SMPD(Neurcomputing2022), IDADA(TIM2022), YOLOv3-promote(TITS2022), HQPFG(PR2022), SF-UPD(ICME2022), IFDNN(TITS2022), SWDR-SR(TIM2022), Region NMS(Neurocomputing2022), IPOPD(arXiv2022); MPD: DCMNet(ACMMM2022), MICNN(CVIU2022), LG-FAPF(IF2022), UTVDA(PRL2022); Others: FSSPA(CVPR2022W)
• [May 10, 2022]: PD: OAF-Net(TITS2022), F2DNet(ICPR2022), Pedstron(arXiv2022); MPD: CMPD(TMM2022); Others: STCrowd(CVPR2022),
• [Mar. 4, 2022]: PD: DMSFLN(TITS2021), SA-DPM(TITS2022), SSC(TITS2022), CFL(TITS2022); MPD: UGCML(TCSVT2021), [MuFEm(TITS2022)], BAANet(arXiv2021); Others: UAVPed(TMM2021), RAHD(TMM2022)
• [Dec. 5, 2021]: PD: EGCL(arXiv2021), AutoPedestrian(TIP2021), SADet(IJCB2021), PAMS-FCN(TITS2021), SAN(TIP2021), Un2Reliab(TMM2021)
• [Nov. 1, 2021]: PD: CRML(ICCV2021); MPD: GAFF(WACV2021)
• [Aug. 31, 2021]: Dataset: MOTSynth(ICCV2021), LLVIP(ICCVW2021); MPD: MRMIoU(MVA2021)
• [July 4, 2021]: PD: NMS-Loss(ICMR2021), VPD(arXiv2021); MPD: SCDN(arXiv2021); Others: SBBG(CVIU2021)
• [April 16, 2021]: PD: LLA(arXiv2021), Box Re-Ranking(arXiv2021), V2F-Net(arXiv2021)
• [Mar. 19, 2021]: PD: DRNet(arXiv2021)
• [Jan. 07, 2021]: PD: DETR for Pedestrian Detection(arXiv2020)
• [Dec. 05, 2020]: PD: KGSNet(TNNLS2020), SSAM(TITS2020), MGAN(TIP2020), PEN(TITS2020), RSA-YOLO(TIP2020), CWETM(TVT2020), PLM(TITS2020), GRPN(TITS2020); Others: SSD-MR(ICRA2020), ADGN(TITS2020)
• [Nov. 19, 2020]: Dataset: A newly built deverse pedestrian detection dataset: TJU-DHD-Pedestrian(TIP2020)
• [Nov. 05, 2020]: PD: TinyCityPersons(WACV2021)
• [Oct. 22, 2020]: PD: BGCNet(ACM-MM2020), NOH-NMS(ACM-MM2020), SML(ACM-MM2020), HGPD(ACM-MM2020)
• [Oct. 07, 2020]: Comparison of multispectral pedestrian detection in leaderboard
• [Oct. 01, 2020]: PD: PRNet(ECCV2020), Case(ECCV2020); MPD: MBNet(ECCV2020); Others: TCDet(ECCV2020);

Table of Contents

1. Detection pipeline
   1.1 Proposal generation
   1.2 Feature extraction
   1.3 Proposal classification
   1.4 Post processing
2. Single-spectral pedestrian detection
   2.1 Handcrafted features based pedestrian detection
   2.1.1 Channel features based methods
   2.1.2 Deformable part based methods
   2.2 Deep features based pedestrian detection
   2.2.1 Hybrid methods
   2.2.2 Pure CNN based methods
3. Multispectral pedestrian detection
   3.1 Deep feature fusion
   3.2 Data processing
   3.3 Domain adaptation
4. Datasets
   4.1 Earlier pedestrian datasets
   4.2 Modern pedestrian datasets
   4.3 Multispectral pedestrian datasets
5. Challenges
   5.1 Scale variance
   5.2 Occlusion
   5.3 Domain adaptation
6. Related Survey