| 摘 要: 针对密集场景中行人遮挡和尺度较小导致的漏检问题,提出了一种基于 YOLOv8(You Only Look Once version 8)的改进行人检测算法。首先,使用 GhostConv(Ghost Convolution)代替部分Conv,减少模型的参数量;其次,融合GSConv(Group Shuffled Convolution)模块,在降低模型复杂度的同时保持精度;最后,引入SNs(Slim Necksolution)模块,解决行人遮挡等原因导致的漏检问题,提升了行人检测精度。将该算法应用于 WiderPerson(行人检测数据集),实验结果表明,相较于原始YOLOv8算法,其在 mAP@0.5指标上提升了2.5百分点,同时在更为严格的 mAP@0.5:0.95指标上也实现了2.1百分点的精度提升。改进的算法可以较好地应用于拥挤行人检测任务。 |
| 关键词: 深度学习;目标检测;行人检测;YOLOv8;轻量化改进 |
|
中图分类号: TP391.4
文献标识码: A
|
|
| The Improved Pedestrian Detection Algorithm Based on YOLOv8 |
|
JIANG Nianzu, HU Min, ZHANG Wei, LIN Shuainan, ZHAO Rui
|
(College of Mathematics and Computer Science, Jilin Normal University, Siping 136000, China)
1355035318@qq.com; 473475826@qq.com; wzhang@sina.cn; 9547133319@qq.com; lijiatong_zr@163.com
|
| Abstract: To address the missed detection issues caused by pedestrian occlusion and smal-l scale targets in dense scenarios, this paper proposes an improved pedestrian detection algorithm based on YOLOv8 (You Only Look Once version 8). Firstly, GhostConv (Ghost Convolution) is employed to replace partial Conv layers, effectively reducing the parameter count of the model. Secondly, the GSConv (Group Shuffled Convolution) module is integrated to reduce computational complexity while maintaining detection accuracy. Finally, the SNs (Slim-Neck solution) module is introduced to mitigate detection failures caused by pedestrian occlusion, thereby improving the overall detection performance. Results of experimental validation on the WiderPerson dataset ( a pedestrian detection benchmark) demonstrate that compared with the original YOLOv8 algorithm, the improved method achieves an improvement of 2.5 percentage points in mAP@0.5 metric and also realizes an increase of 2.1 percentage points in the more stringent mAP @0.5:0.95 metric. The improved algorithm can be effectively applied to crowded pedestrian detection tasks. |
| Keywords: deep learning; object detection; pedestrian detection; YOLOv8; lightweight optimization |
