基于深度学习的人体行为识别
Behavior Recognition of Human Based on Deep Learning
-
摘要: 为了识别公共区域等特定场所下的人体行为,提出了一种基于深度学习的人体行为识别方法。首先,预处理训练样本集和测试样本集中的所有图像,通过高斯混合模型提取出目标运动前景。其次,对训练样本集中各种目标行为建立样本库,定义不同类别的识别行为作为先验知识,用于训练深度学习网络。最后,结合深度学习所得到的网络模型,分类识别测试样本集中的各种行为,并将识别的结果和当前流行方法进行了比较。实验结果表明,该人体行为识别方法优于其它方法,平均识别率相比其他方法有较大的提高。Abstract: To recognize human behaviors in public areas, a new method of recognition was proposed based on deep learning. First, we pre-processed all the images in training and test samples, and utilized GMM to extract moving objects. Then, we built sample sets of various behaviors, and defined different behaviors as priori knowledge to train a deep learning network. Finally, all kinds of behaviors based on the network model of deep learning were recognized. Experimental results demonstrated our method outperforms the existed methods, and the average recognition rate is 96.82%.
下载:
