基于联合学习的多视角室内人员检测网络

Multi-View Indoor Human Detection Neural Network Based on Joint Learning

建立了室内人员检测数据集(IHDD),提出了基于联合学习的多视角室内人员检测网络模型(MVNN)。该模型由输入数据层、特征提取层、可变形处理层、可见性估计层、分类判别层等组成,并加入区域建议模型和多视角模型以提升算法的检测性能。在自建的IHDD数据集上的实验结果表明,与现有其他检测算法相比,MVNN算法的检测率更高;在人体目标呈现多视角、多姿态及存在遮挡等困难情况下仍有不错的检测效果,具有一定的理论研究价值和实际应用价值。

An indoor human detection dataset(IHDD) is established, and a novel multi-view indoor human detection neural network(MVNN) based on joint learning is proposed. The model consists of input data layer, feature extraction layer, deformation layer, visibility reasoning layer and classification layer, and the proposed MVNN algorithm can improve the detection performance when combined with the region proposal model and the multi-view model. Experimental results on the self-built IHDD show that compared with other existing detection algorithms, the proposed MVNN algorithm has a higher detection rate. It can still obtain good detection results even in the case of difficult situations such as various views, changing poses and occlusion for human targets, which indicates certain theoretical research value and practical value.