Lune

CVPR Findings2026

WideEye: Achieving Wide Field-of-view Traffic Video Analytics With Dynamic Orientation Adaptation

Z. Jonny Kong, Sibendu Paul, Y. Charlie Hu

Abstract

Wide field-of-view (WFoV) video analytics is crucial for applications such as traffic monitoring and surveillance, where objects of interest appear across large spatial areas. While 360 • or fisheye cameras can provide full coverage, their high cost and limited image quality hinder widespread use. In contrast, low-cost pan-tilt-zoom (PTZ) cameras are already extensively deployed and can mechanically rotate to cover wide scenes over time. However, existing video analytics methods either assume static cameras or optimize accuracy only within the currently visible field-of-view (FoV), failing to achieve accurate analytics across the entire WFoV.

We present WideEye, a learning-augmented video analytics system that enables high-accuracy WFoV analytics using a single PTZ camera through dynamic orientation adaptation. WideEye introduces two key components: (1) a trajectory prediction model, trained per scene to infer object motion across intermittently observed FoVs based on past detection results, and (2) a model-predictive FoV scheduler that learns to select the next orientation by jointly reasoning about scene visibility and dynamics that affect prediction reliability.

Experiments on real traffic videos using commodity PTZ cameras demonstrate that WideEye substantially improves wide-view object detection accuracy, narrowing the mAP gap between PTZ and true WFoV setups from up to 30.6% to only 10.0%. These results highlight that coupling learningbased motion modeling with predictive camera control can transform conventional PTZ cameras into effective widearea video analytics sensors.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

Questions to start from

Lune can draw on

the full paper · related work

Get the answer

Free to start. No credit card required.

Lineage

  • Gemel: Model Merging for Memory-Efficient, Real-Time Video Analytics at the Edge cites Server-Driven Video Streaming for Deep Learning Inference
Builds onCited byLeft to right by year; the more cited, the higher.

Builds on13

  1. Server-Driven Video Streaming for Deep Learning InferenceKuntai Du, Ahsan Pervaiz, Xin Yuan, Aakanksha Chowdhery et al.SIGCOMM 2020 · 238 citations
  2. Gemel: Model Merging for Memory-Efficient, Real-Time Video Analytics at the EdgeArthi Padmanabhan, Neil Agarwal, Anand P. Iyer, Ganesh Ananthanarayanan et al.NSDI 2023 · 94 citations
  3. Edge-assisted Online On-device Object Detection for Real-time Video AnalyticsMengxi Hanyao, Yibo Jin, Zhuzhong Qian, Sheng Zhang et al.INFOCOM 2021 · 82 citations
  4. Large-scale vehicle trajectory reconstruction with camera sensing networkPanrong Tong, Mingqian Li, Mo Li, Jianqiang Huang et al.MobiCom 2021 · 58 citations
  5. EdgeDuet: Tiling Small Object Detection for Edge Assisted Autonomous Mobile VisionXu Wang, Zheng Yang, Jiahang Wu, Yi Zhao et al.INFOCOM 2021 · 54 citations

Related papers

  1. MadEye: Boosting Live Video Analytics Accuracy with Adaptive Camera ConfigurationsMike Wong, Murali Ramanujam, Guha Balakrishnan, Ravi NetravaliNSDI 2024 · 17 citations
  2. Uirapuru: Timely Video Analytics for High-Resolution Steerable Cameras on Edge DevicesGuilherme Henrique Apostolo, Pablo Bauszat, Vinod Nigade, Henri E. Bal et al.MobiCom 2025 · 1 citation
  3. Cross-Camera Inference on the Constrained EdgeJingzong Li, Libin Liu, Hong Xu, Shudeng Wu et al.INFOCOM 2023 · 31 citations
  4. OmniSense: Towards Edge-Assisted Online Analytics for 360-Degree VideosMiao Zhang, Yifei Zhu, Linfeng Shen, Fangxin Wang et al.INFOCOM 2023 · 8 citations
  5. WiseCam: Wisely Tuning Wireless Pan-Tilt Cameras for Cost-Effective Moving Object TrackingJinlong E, Lin He, Zhenhua Li, Yunhao LiuINFOCOM 2023 · 4 citations
Dusk over the sea between two cliffs drawn in fine vertical lines