首页|基于机器视觉的运动目标跟踪算法优化研究

基于机器视觉的运动目标跟踪算法优化研究

扫码查看
针对全卷积孪生神经网络目标跟踪速度较慢等问题,研究基于全卷积孪生网络(fully-convolutional siamese networks,SiamFC)算法提出一种卷积层替换的改进方法.将原算法的卷积层进行相应的替换,减少卷积过程中的参数计算量,以此提升网络速度以及判断力.研究采用一种深度可分离卷积与混合深度卷积进行结构更新,通过不同尺寸卷积核提取图像特征.研究利用数据集对改进算法进行验证,实验中,改进SiamFC算法相较于SiamFC算法的一次性评估成功率提了 2%左右,精度提升了 3%左右;在空间鲁棒性上的成功率提升了 3%左右,精度提升4%左右;在时间鲁棒性上的成功率分别提升2%左右,精度分别提升5%左右.结果表明,研究改进的算法的性能得到一定程度提升.
Optimization of Moving Object Tracking Algorithm Based on Machine Vision
In response to the problem of slow target tracking speed in fully convolutional twin neural networks,an improved method of convolutional layer replacement is proposed based on the fully convolutional twin networks(SiamFC)algorithm.Replace the convolutional layer of the original algo-rithm accordingly to reduce the calculation of parameters during the convolution process,thereby impro-ving network speed and judgment.The research adopts a deep separable convolution and a mixed deep convolution for structural updates,and extracts image features through convolution kernels of different sizes.The study used a dataset to validate the improved algorithm.In the experiment,the one-time e-valuation success rate of the improved SiamFC algorithm was improved by about 2%compared to the SiamFC algorithm,and the accuracy was improved by about 3%.The success rate in spatial robustness has been improved by about 3%,and the accuracy has been improved by about 4%.The success rate in terms of time robustness has been improved by about 2%,and the accuracy has been improved by about 5%.The results indicate that the performance of the improved algorithm has been improved to a certain extent.

machine visionconvolutionaltarget trackingfeature extractionSiamFC

周小勇

展开 >

漳州职业技术学院 福建漳州 363000

机器视觉 卷积 目标跟踪 特征提取 SiamFC

2024

佳木斯大学学报(自然科学版)
佳木斯大学

佳木斯大学学报(自然科学版)

影响因子:0.159
ISSN:1008-1402
年,卷(期):2024.42(2)
  • 12