Abstract: The object detection algorithm based on deep learning cannot achieve good detection speed without the support of high-performance GPU hardware devices. However, the hardware equipment with high performance, high power consump-tion and large size in the intelligent car does not conform to the concept of long endurance of the car. Therefore, this paper takes YOLOV3 object detection algorithm as the baseline model for improvement, proposes a lightweight object detection model, Mobile-YOLO, and conducts training and testing in the collected and produced underground parking lot dataset. The experimental results show that compared with Yolov3, the proposed Mobile-YOLO has a 47.1% increase in detection speed with a slight increase in the mean accuracy. It can detect 31 images per second on the mobile terminal platform TX2.

Key words: Deep learning; Object detection; Lightweight; Mobile terminal

摘要： 基于深度学习的目标检测算法能够取得良好的检测速度离不开高性能 GPU 硬件设备的支持。 然而，在智能车中搭载高性能、高功耗、大尺寸的硬件设备与汽车的长续航理念不符。因此，文 章以 YOLOv3 目标检测算法为基线模型进行改进，提出轻量化的目标检测模型 Mobile-YOLO，并 在采集制作的地下停车场数据集中进行训练测试。实验结果表明，提出了 Mobile-YOLO 相较于 YOLOv3，在平均精度均值略微提升的情况下，检测速度提升了 47.1%。在移动端平台 TX2 上每 秒能够检测 31 张图像。

关键词: 深度学习；目标检测；轻量化；移动端