【摘要】 通过比较分析视频和图像两种数据集,参考白、拉普拉斯算子等多种图像处理方法,归一化RGB模型和YCrCb模型两种肤色提取方法,全图像像素、连通域质心距离和双手(臂)质心坐标三种应用特征提取方法,KNN、决策树、神经网络等七种姿态分类方法,比较研究驾驶员姿态识别过程中各方法的优劣性及适用范围.分析结果表明:驾驶员姿态识别离形成成熟的产品还有一定的距离,其主要原因在于肤色模型抗光线变化干扰、姿态识别精度和实时性方面尚未达到商业应用的要求.更多还原

【Abstract】 Through comparative analysis of video and image data sets, referring to various image processing methods such as white and Laplacian operators, two skin color extraction methods including normalized RGB model and YCrCb model, three applied feature extraction methods such as full image pixels, connected domain centroid distance and two-hand(arm) centroid coordinates, and seven attitude classification methods such as KNN, decision tree and neural network, the advantages and disadvantages and applicable scope of each method in the process of driver attitude identification were compared and studied.The analysis results also show that there is still a certain distance between driver attitude recognition and mature products. The main reason is that the skin color model has not met the requirements of commercial application in terms of anti-interference of light changes, accuracy and real-time performance of attitude recognition.更多还原