Omnidirectional Pedestrian Detection by Rotation Invariant Training

Tamura, Masato; Horiguchi, Shota; Murakami, Tomokazu

doi:10.1109/wacv.2019.00216

Cited by 44 publications

(40 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Considering the attractive performance and straightforward implementation, most existing omnidirectional pedestrian detection methods (Nguyen et al, 2016;Seidel et al, 2019;Li et al, 2019;Duan et al, 2020), as well as ours in the conference paper (Tamura et al, 2019), adopt YOLO as a base detector. YOLO uses pixel-wise dense prediction, which outputs duplicated detection results, and thus requires a suppression postprocess.…”

Section: Object Detectionmentioning

confidence: 99%

“…The default data augmentation is used in this type. We consider this detector as a baseline because this training method is that of our conference paper (Tamura et al, 2019) enhanced with the angle-aware detection.…”

Section: Quantitative Comparisonsmentioning

confidence: 99%

“…As a sub-problem of pedestrian detection, omnidirectional pedestrian detection has also been studied for years (Cinaroglu and Bastanlar, 2014;Krams and Kiryati, 2017;Demirkus et al, 2017;Nguyen et al, 2016;Seidel et al, 2019;Tamura et al, 2019;Li et al, 2019;Duan et al, 2020). A major difference between omnidirectional pedestrian detection and the general one is that the appearance of pedestrians tends to be deformed due to the wide angle of view by omnidirectional cameras.…”

Section: Introductionmentioning

confidence: 99%

“…To address the aforementioned problems of the existing methods, we propose a training method that fully leverages an existing large-scale object detection dataset. In our previous conference paper (Tamura et al, 2019), we focused on the rotation of the appearance in omnidirectional images. To deal with the rotation without transformation at inference time, we applied rotation data augmentation to images in the COCO object detection dataset (Lin et al, 2014) during training.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Segmentation-Based Bounding Box Generation for Omnidirectional Pedestrian Detection

Tamura¹,

Yoshinaga²

2021

Preprint

Self Cite

View full text Add to dashboard Cite

We propose a segmentation-based bounding box generation method for omnidirectional pedestrian detection, which enables detectors to tightly fit bounding boxes to pedestrians without omnidirectional images for training. Because the appearance of pedestrians in omnidirectional images may be rotated to any angle, the performance of common pedestrian detectors is likely to be substantially degraded. Existing methods mitigate this issue by transforming images during inference or training detectors with omnidirectional images. However, the first approach substantially degrades the inference speed, and the second approach requires laborious annotations. To overcome these drawbacks, we leverage an existing large-scale dataset, whose segmentation annotations can be utilized, to generate tightly fitted bounding box annotations. We also develop a pseudo-fisheye distortion augmentation method, which further enhances the performance. Extensive analysis shows that our detector successfully fits bounding boxes to pedestrians and demonstrates substantial performance improvement.

show abstract

Section: Object Detectionmentioning

confidence: 99%

Section: Quantitative Comparisonsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Segmentation-Based Bounding Box Generation for Omnidirectional Pedestrian Detection

Tamura¹,

Yoshinaga²

2021

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…However, the transformation relies heavily on calibrated camera parameters, which requires user-interaction. Other works [29,22] train orientation-aware networks using pedestrian images with synthetic rotation. However, such training introduces large computational cost and the trained network is biased towards the data used.…”

Section: Omnidirectional Video Analysismentioning

confidence: 99%

Weakly-Supervised Multi-Person Action Recognition in 360° Videos

Liu²,

Wang

et al. 2020

2020 IEEE Winter Conference on Applications of Computer Vision (WACV)

View full text Add to dashboard Cite

The recent development of commodity 360 • cameras have enabled a single video to capture an entire scene, which endows promising potentials in surveillance scenarios. However, research in omnidirectional video analysis has lagged behind the hardware advances. In this work, we address the important problem of action recognition in topview 360 • videos. Due to the wide filed-of-view, 360 • videos usually capture multiple people performing actions at the same time. Furthermore, the appearance of people are deformed. The proposed framework first transforms top-view omnidirectional videos into panoramic videos using a calibration free method. Then spatial-temporal features are extracted using region-based 3D CNNs for action recognition. We propose a weakly-supervised method based on multiinstance multi-label learning, which trains the model to recognize and localize multiple actions in a video using only video-level action labels as supervision. We perform experiments to quantitatively validate the efficacy of the proposed method over state-of-the-art baselines and variants of our model, and qualitatively demonstrate action localization results. To enable research in this direction, we introduce the 360Action dataset. It is the first omnidirectional video dataset for multi-person action recognition with a diverse set of scenes, actors and actions. The dataset is available at https://github.com/ryukenzen/360action. * This work is mainly completed during Junnan Li's period of internship at NEC Corporation

show abstract