中圖分類號： TP391.4 文獻標(biāo)識碼： A DOI：10.16157/j.issn.0258-7998.212121 中文引用格式： 王均成，賀超,，趙志源,，等. 基于YOLOv3-tiny的視頻監(jiān)控目標(biāo)檢測算法[J].電子技術(shù)應(yīng)用，2022,，48(7)：30-33,，39. 英文引用格式： Wang Juncheng，He Chao,，Zhao Zhiyuan,，et al. Video surveillance object detection method based on YOLOv3-tiny[J]. Application of Electronic Technique，2022,，48(7)：30-33,，39.
Video surveillance object detection method based on YOLOv3-tiny
Wang Juncheng1，2,，3,，He Chao1，2,，3,，Zhao Zhiyuan1，2,，3,，Zou Jianwen1,，2，3
1.School of Communication and Information Engineering,，Chongqing University of Posts and Telecommunications,， Chongqing 400065，China,； 2.Advanced Network and Intelligent Connection Technology Key Laboratory of Chongqing Education Commission of China,， Chongqing 400065，China,； 3.Chongqing Key Laboratory of Ubiquitous Sensing and Networking,，Chongqing 400065，China
Abstract： Object detection methods have great value in the application field of video surveillance. At present, it is difficult to realize real-time object detection in resource constrained video surveillance system. A object detection method based on improved YOLOv3-tiny is proposed. Based on the YOLOv3-tiny architecture, the algorithm optimizes the backbone network by adding feature reuse, and a fully-connected attention mix module is proposed to enable the network to learn more abundant spatial information, which is more suitable for object detection under resource constraints. The experimental data shows that compared with YOLOv3-tiny, the algorithm reduces the model volume by 39.2%, the amount of parameters by 39.8%, and improves the mAP of 2.7% on the VOC data set, which significantly reduces the occupation of model resources while improving the detection accuracy.
Key words : object detection,；video surveillance,；YOLOv3；feature reuse,；attention mechanism
為了實現(xiàn)資源有限設(shè)備上目標(biāo)檢測這一挑戰(zhàn),，人們對研究和設(shè)計低復(fù)雜度的高效神經(jīng)網(wǎng)絡(luò)體系架構(gòu)越來越感興趣。而著名的YOLO[5](You Only Look Once,，YOLO)則是圍繞效率設(shè)計的一階段目標(biāo)檢測算法,，它可以在高端圖形處理器上實現(xiàn)視頻監(jiān)控目標(biāo)高效檢測。然而對于許多資源受限監(jiān)控設(shè)備來說,，這些網(wǎng)絡(luò)架構(gòu)參數(shù)量大且計算復(fù)雜度較高,，使得在嵌入式等監(jiān)控設(shè)備上運行時推理速度大幅下降。YOLOv3[6]是YOLO系列應(yīng)用在各領(lǐng)域最普遍的算法,，YOLOv3-tiny則是在該算法的基礎(chǔ)上簡化的,，雖然精度顯著下降但具有了更少計算成本,，這大大增加了在資源受限監(jiān)控設(shè)備上部署目標(biāo)檢測算法的可行性。
