YOLO-wLU：Object Detection Algorithm Considering Localization Uncertainty

doi:10.3778/j.issn.1002-8331.2007-0234

Abstract

Abstract:

The standard object detection algorithm, YOLOv3, does not consider the bounding box coordinates localization uncertainty. Consequently, it will yield incorrect detection results sometimes. To solve this problem, the YOLO-wLU algorithm（YOLO with Localization Uncertainty） based on YOLOv3 is proposed. By the idea of uncertainty in the deep learning, the Gaussian distribution function is used for modeling the bounding box coordinates probability distribution to consider the bounding box coordinates localization uncertainty. Moreover, a new bounding box loss function is designed, and the detection results with larger localization uncertainty is removed. Meanwhile, the accuracy of bounding box coordinates identification results is improved by merging the surrounding bounding boxes coordinates information. The experimental results show that the proposed YOLO-wLU algorithm effectively reduces the FP（False Positive） rate, and improves the detection precision; more concretely, the proposed YOLO-wLU algorithm improves mAP at most 4.1 percentage points compared with the YOLOv3 algorithm on the COCO test dataset.

Key words: object detection, YOLOv3 algorithm, YOLO-wLU algorithm, localization uncertainty

摘要：

YOLOv3目标检测算法在检测目标时没有考虑边界框坐标定位存在的不确定性，因此有时不能得出正确的检测结果。针对此问题，提出YOLO-wLU（YOLO with Localization Uncertainty）算法。该算法借鉴深度学习中的不确定性思想，使用高斯分布函数建立边界框坐标的概率分布模型以考虑边界框坐标定位不确定性；设计新的边界框损失函数，在检测过程中移除定位不确定性较大的检测结果；通过融合周围边界框坐标信息提高了边界框坐标辨识结果的准确性。实验结果表明，该算法可有效减少误报率，提高检测精度；COCO数据集上测试结果显示，相比YOLOv3算法，该算法的mAP最高可提升4.1个百分点。

关键词: 目标检测, YOLOv3算法, YOLO-wLU算法, 定位不确定性

XIE Xiong, YANG Jinpeng. YOLO-wLU：Object Detection Algorithm Considering Localization Uncertainty[J]. Computer Engineering and Applications, 2021, 57(22): 223-231.

谢兄，杨金鹏. YOLO-wLU：考虑定位不确定性的目标检测算法[J]. 计算机工程与应用, 2021, 57(22): 223-231.

References

[1] LECUN Y，BOSER B，DENKER J S，et al.Backpropagation applied to handwritten zip code recognition[J].Neural Computation，1989，1（4）：541-551.
[2] VAILLANT R，MONROCQ C，LECUN Y.Original approach for the localisation of objects in images[J].IEE Proceedings-Vision Image and Signal Processing，1994，141（4）：245-250.
[3] GIRSHICK R，DONAHUE J，DARRELL T，et al.Rich feature hierarchies for accurate object detection and semantic segmentation[C]//2014 IEEE Conference on Computer Vision and Pattern Recognition，2014：580-587.
[4] HE K，ZHANG X，REN S，et al.Spatial pyramid pooling in deep convolutional networks for visual recognition[J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2014，37（9）：1904-1916.
[5] GIRSHICK R.Fast R-CNN[C]//2015 IEEE International Conference on Computer Vision，2015：1440-1448.
[6] REN S，HE K，GIRSHICK R，et al.Faster R-CNN：towards real-time object detection with region proposal networks[J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2015，39（6）：1137-1149.
[7] HE K M，GEORGIA G，PIOTR D，et al.Mask R-CNN[J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2020，42（2）：386-397.
[8] REDMON J，DIVVALA S，GIRSHICK R，et al.You only look once：unified，real-time object detection[C]//2016 IEEE Conferenceon Computer Vision and Pattern Recognition，2016：779-788.
[9] REDMON J，FARHADI A.YOLO9000：better，faster，stronger[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition，2017：6517-6525.
[10] REDMON J，FARHADI A.YOLOv3：an incremental improvement[J].arXiv：1804.02767，2018.
[11] LIU W，ANGUELOV D，ERHAN D，et al.SSD：single shot multibox detector[C]//14th European Conference on Computer Vision，2016：21-37.
[12] FU C Y，LIU W，RANGA A，et al.DSSD：deconvolutional single shot detector[J].arXiv：1701.06659，2017.
[13] SHEN Z，LIU Z，LI J，et al.DSOD：learning deeply supervised object detectors from scratch[C]//2017 IEEE International Conference on Computer Vision，2017：1919-1927.
[14] LIN T Y，GOYAL P，GIRSHICK R，et al.Focal loss for dense object detection[J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2020，42（2）：318-327.
[15] 徐诚极，王晓峰，杨亚东.Attention-YOLO：引入注意力机制的YOLO检测算法[J].计算机工程与应用，2019，55（6）：13-23.
XU C J，WANG X F，YANG Y D.Attention-YOLO：YOLO detection algorithm that introduces attention mechanism[J].Computer Engineering and Applications，2019，55（6）：13-23.
[16] 温捷文，战荫伟，凌伟林，等.实时目标检测算法YOLO的批再规范化处理[J].计算机应用研究，2018，35（10）：3179-3185.
WEN J W，ZHAN Y W，LING W L，et al.Batch re-normalization of real-time object detection algorithm YOLO[J].Application Research of Computers，2018，35（10）：3179-3185.
[17] KENDALL A，GAL Y.What uncertainties do we need in Bayesian deep learning for computer vision?[C]//2017 Annual Conference on Neural Information Processing Systems，2017：5574-5584.
[18] FENG D，ROSENBAUM L，DIETMAYER K.Towards safe autonomous driving：capture uncertainty in the deep neural network for lidar 3D vehicle detection[C]//2018 International Conference on Intelligent Transportation Systems，2018：3266-3273.
[19] FENG D，ROSENBAUM L，TIMM F，et al.Leveraging heteroscedastic aleatoric uncertainties for robust real-time LIDAR 3D object detection[J].arXiv：1809.05590，2018.
[20] HE Y，ZHANG X，SAVVIDES M，et al.Softer-NMS：rethinking bounding box regression for accurate object detection[J].arXiv：1809.08545，2018.
[21] CHOI J，CHUN D，KIM H，et al.Gaussian YOLOv3：an accurate and fast object detector using localization uncertainty for autonomous driving[C]//2019 IEEE International Conference on Computer Vision，2019：502-511.
[22] BLAKE A，CURWEN R，ZISSERMAN A.A framework for spatiotemporal control in the tracking of visual contours[J].International Journal of Computer Vision，1993，11（2）：127-145.
[23] HE X，ZEMEL R S，CARREIRA-PERPINAN M A.Multiscale conditional random fields for image labeling[C]//2004 IEEE Conference on Computer Vision and Pattern Recognition，2004：695-703.
[24] GAL Y.Uncertainty in deep learning[D].Cambridge：University of Cambridge，2016.
[25] HE K，ZHANG X，REN S，et al.Deep residual learning for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition，2016：770-778.
[26] NEUBECK A，GOOL L J V.Efficient non-maximum suppression[C]//18th International Conference on Pattern Recognition，2006：850-855.