| 基于特征融合的复杂场景多目标跟踪算法研究 |
| |
| 引用本文： | 王志余.基于特征融合的复杂场景多目标跟踪算法研究[J].教育技术导刊,2020,19(4):46-49. |
| |
| 作者姓名： | 王志余 |
| |
| 作者单位： | 1. 山东科技大学 计算机学院，山东 青岛 266590；2. 黄岛中医医院，山东 青岛 266500 |
| |
| 基金项目： | 山东省自然科学基金项目（ZR2018BF001） |
| |
| 摘 要： | 在复杂地物类型背景条件下，多目标跟踪算法通常表现出目标识别与跟踪能力较差问题，特别在被其它地物遮挡后目标跟踪丢失更严重。提出一种改进的基于多源特征提取与特征融合的多目标跟踪算法。为提高目标在复杂背景下的空间分辨力，充分利用对异类物体判别能力较强的高层特征和针对同类不同物体判别能力较强的浅层特征，提高复杂背景下地物目标的识别能力。同时，为了解决物体被遮挡后导致跟踪算法丢失目标问题，利用滤波器获得追踪目标的空间尺度大小，提高跟踪算法的准确性与可靠性。实验表明，多目标跟踪算法识别目标的准确性可达87.5%，误差在±2.31%]左右，具有良好的尺度估计效果。
|
| 关 键 词： | 机器视觉 特征融合 目标跟踪 尺度估计 运动检测 |
| 收稿时间： | 2019-04-25 |
Research on Multi-target Tracking Algorithm for Complex Background Based on Feature Fusion |
| |
| WANG Zhi-yu.Research on Multi-target Tracking Algorithm for Complex Background Based on Feature Fusion[J].Introduction of Educational Technology,2020,19(4):46-49. |
| |
| Authors: | WANG Zhi-yu |
| |
| Institution: | 1. College of Computer Science and Engineering， Shandong University of Science and Technology， Qingdao 266590， China；2. Huangdao District Chinese Medicine Hospital， Qingdao 266500， China |
| |
| Abstract: | In the context of complex object types， multi-target tracking algorithms usually show poor target recognition and target tracking ability， especially after being occluded by other objects， which will lead to the loss of tracking set targets. A multi-target tracking algorithm for multi-source feature extraction and feature fusion is proposed in order to improve the spatial resolving power of neural networks， make full use of different depth features and high-level features that have strong discriminative ability for heterogeneous objects and discriminate against different objects of the same kind. Strong shallow features and multi-source fusion for features are fully used to improve the recognition of ground objects. At the same time， in order to solve the problem that the tracking algorithm loses the target after the object is occluded， the scale filter is used to calculate the spatial scale of the target， and the performance of the multi-target tracking algorithm is improved. Field experiments show that multi-target tracking algorithm can achieve target accuracy of 87.5%， error rate is±2.31%]， and the algorithm has good scale estimation effect. |
| |
| Keywords: | machine vision feature fusion target tracking scale estimation motion detection |
|
| 点击此处可从《教育技术导刊》浏览原始摘要信息 |
| 点击此处可从《教育技术导刊》下载免费的PDF全文 |
