Preprocessing via Deep Learning for Enhancing Real-Time Performance of Object Detection
Preprocessing via Deep Learning for Enhancing Real-Time Performance of Object Detection
复制标题
DOI：
10.1109/vtc2023-spring57618.2023.10200997
复制
发表时间：
2023-06
期刊：
影响因子：
--
通讯作者：
Yu Liu;K. Kang
中科院分区：
文献类型：
--
作者：
Yu Liu;K. Kang
深度学习模型显著改进了交通监控中必不可少的目标检测。然而，这些模型的日益复杂导致了更高的延迟和资源消耗，使得实时目标检测具有挑战性。为了解决这个问题，我们提出了一个轻量级的深度学习模型，称为空路检测(ERD)。ERD通过二值分类有效地识别并去除不包含任何感兴趣对象的空交通图像，例如车辆。通过充当预处理单元，ERD过滤掉不重要的数据，从而降低了计算复杂性和延迟。ERD高度兼容，可以与任何第三方物体检测模型无缝合作。在我们的评估中，我们发现对于真实的交通监控视频，ERD将EfficientDet、SSD和YOLOV5的帧处理率分别提高了约44%、40%和10%。
Deep learning models have significantly improved object detection essential for traffic monitoring. However, these models’ increasing complexity results in higher latency and resource consumption, making real-time object detection challenging. To address this issue, we propose a lightweight deep learning model called Empty Road Detection (ERD). ERD efficiently identifies and removes empty traffic images that do not contain any object of interest, such as vehicles, via binary classification. By serving as a preprocessing unit, ERD filters out nonessential data, reducing computational complexity and latency. ERD is highly compatible and can work seamlessly with any third-party object detection model. In our evaluation, we found that ERD improves the frame processing rate of EfficientDet, SSD, and YOLOV5 by approximately 44%, 40%, and 10%, respectively, for a real-world traffic monitoring video.