Lightweight Pedestrian Detection Algorithm Combined with Data Enhancement

Lifen Li ； Minghe Ma

《World Scientific Research Journal》 8卷5期 (2022/05) Pp. 201-210

https://doi.org/10.6911/WSRJ.202205_8(5).0026

摘要

YOLOv3 has become a commonly used target detection algorithm in the industrial field due to its fast detection speed and high detection accuracy. But the disadvantage is that the network model is too large to be easily deployed on small terminals. In order to solve the problem of YOLOv3 big network model and further improve its detection speed and detection accuracy, a lightweight pedestrian detection algorithm combined with data enhancement is proposed. Through using bilinear interpolation, Mosaic data enhancement and other methods to optimize the pedestrian image, and use the lightweight network MobileNetV3 to replace the backbone network of YOLOv3, and select the Kmeans++ algorithm to replace the original Kmeans algorithm used to obtain the anchor boxes. Experiment on the current mainstream pedestrian data set. Experimental results show that the size of the improved algorithm model is 1/20 of the original YOLOV3 algorithm model, while the detection speed is increased to 91FPS, and the detection accuracy has also been improved to a certain extent.

關鍵字

Pedestrian detection ； Deep learning ； YOLOv3 ； MobileNetV3 ； Mosaic data enhancement

參考文獻

R. Girshick, J. Donahue, T. Darrell and J. Malik. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//2014 IEEE Conference on Computer Vision and Pattern Recognition. 2014:580-587.

R. Girshick. Fast R-CNN[C]//2015 IEEE International Conference on Computer Vision(ICCV).2015:1440-1448

S. Ren, K. He, R. Girshick and J. Sun.Faster R-CNN: towards real-time object detection with region proposal networks[C]//IEEE Transacti-ons on Pattern Analysis and Machine Intelligence.2017,39(6):1137-1149.

J. Redmon, S. Divvala, R. Girshick and A. Farhadi. You only look once: unified, real-time object detection[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).2016:779-788.

J. Redmon and A. Farhadi. YOLO9000: better, faster,stronger[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition(CVPR).2017:6517-6525