基于计算机视觉技术和支持向量机的手势识别算法研究  

Research on Gesture Recognition Algorithm Based on Computer Vision Technology and Support Vector Machinep

作　　者：徐飞[1] 邹寿春 XU Fei;ZOU Shouchun(Information Center of Minxi Vocational & Technical College,Longyan Fujian 364030)

机构地区：[1]闽西职业技术学院信息中心,福建龙岩364030

出　　处：《佳木斯大学学报（自然科学版）》2023年第1期29-33,共5页Journal of Jiamusi University：Natural Science Edition

基　　金：福建省中青年教师教育科研项目资助(JZ181053)。

摘　　要：针对现有的手势识别方法在复杂环境中识别效率不理想的情况,提出一种双通道卷积神经网络模型,该模型同时采用灰度世界算法和离散小波变换对输入数据进行预处理,减少照变化对图像的影响并提高识别效率和模型稳定性。然后通过高维特征融合模块将提取的图像信息进行融合,再利用帝国竞争算法对支持向量机分类器进行优化,提高分类效果。实验结果显示,在实验环境中,该模型的平均识别率达95%,收敛速度快,效率高。经过消融实验对比,性能比基准模型提高4%以上。在实际测试中,对于简单手势的识别率均在90%以上,对于复杂收手势的识别率在80%以上。In view of the unsatisfactory recognition efficiency of existing gesture recognition methods in complex environments, a two channel convolutional neural network model is proposed. The model uses both gray world algorithm and discrete wavelet transform to preprocess input data, reduce the impact of illumination changes on images, and improve recognition efficiency and model stability. Then the extracted image information is fused through the high-dimensional feature fusion module, and then the Empire competition algorithm is used to optimize the support vector machine classifier to improve the classification effect. The experimental results show that in the experimental environment, the average recognition rate of the model is 95%, the convergence speed is fast and the efficiency is high. Compared with the ablation experiment, the performance is improved by more than 4% compared with the benchmark model. In the actual test, the recognition rate of simple gestures is more than 90%, and the recognition rate of complex gestures is more than 80%.

关 键 词：卷积神经网络 灰度图 离散小波变换 帝国竞争算法 支持向量机 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]