基于改進YOLOV7的變規模網絡重疊區域多目標跟蹤方法

摘" 要： 在實際場景中，目標之間常常存在重疊或部分遮擋的情況，若是未進行有效的多目標檢測以及了解變規模網絡重疊區域內的節點狀況，會導致目標跟蹤精度下降。對此，提出一種基于改進YOLOV7的變規模網絡重疊區域多目標跟蹤方法。首先，采用改進YOLOV7對變規模網絡重疊區域多目標進行檢測；然后，在目標檢測的基礎上，對多目標軌跡特征進行提取；最后，基于提取到的多目標軌跡特征，已知目標的速度、方向與距離，實現變規模網絡重疊區域的多目標跟蹤。實驗結果表明，所提方法的跟蹤精準度最高達到98%，曼哈頓距離明顯小于對比方法，僅在0.1～-0.1之間，性能較優，具有實用性。

關鍵詞： 多目標跟蹤； 重疊區域； YOLOV7； 多目標檢測； 軌跡特征提取； 曼哈頓距離

中圖分類號： TN911.23?34； TP391" " " " " " " " " "文獻標識碼： A" " " " " " " " " 文章編號： 1004?373X（2024）12?0057?05

Method of improved YOLOV7 based multitarget tracking for overlapping

regions in variable scale networks

WANG Bo， CHAI Rui

（School of Computer Science and Technology， North University of China， Taiyuan 030051， China）

Abstract： In practical scenarios， there is often overlap or partial occlusion between targets. If effective multitarget detection is not carried out to understand the node status in the overlapping area of the variable scale network， it will lead to a decrease in target tracking accuracy. Therefore， an improved YOLOV7 based multi target tracking method for the overlapping area of the variable scale network is proposed. The improved YOLOV7 is used to detect multiple targets in overlapping areas of the variable scale network. On the basis of target detection， multi target trajectory features are extracted. Multitarget tracking for overlapping areas in the variable scale network is realized based on the extracted multitarget trajectory features， and the given speed， direction， and distance of the targets. The experimental results show that the proposed method has a tracking accuracy of up to 98%， and the Manhattan distance is significantly smaller than that of the comparison method， only between 0.1 and -0.1， which has better performance and practicality.

Keywords： multitarget tracking; overlapping regions; YOLOV7; multitarget detection; trajectory feature extraction; Manhattan distance

0" 引" 言

變規模網絡是指網絡規模在時間和空間上具有動態變化的特點的一種網絡。這種網絡結構可以根據實時需求、資源分配、拓撲結構等因素進行動態調整，以適應不斷變化的環境和需求。隨著人工智能和計算機視覺技術的快速發展，對于高密度目標的準確跟蹤的需求不斷增加[1?3]；且當場景中存在多個目標時，目標之間可能會產生遮擋或重疊現象，傳統方法往往難以準確識別和跟蹤這些目標。而多目標跟蹤算法的準確性和穩定性很大程度上依賴于提取到的數據支撐，如目標的軌跡特征、相似性信息和位置關系等。因此，研究變規模網絡重疊區域多目標跟蹤方法具有重要意義[4?5]。……