基于泛在信号融合的室内外场景鲁棒感知算法

针对室内外场景结合的导航应用服务需求的发展以及现有室内外场景感知方法的识别稳定性较低、准确率不高问题,本文提出一种基于泛在信号融合的室内外场景鲁棒感知算法,利用室内外场景中融合的泛在信号降低单一信号识别误差;同时为提高传统AdaBoost算法对不平衡数据集的分类精度,采用概率神经网络（PNN）作为训练的弱分类器,并引入熵权法,对迭代产生的弱分类器的权重进行修正来提高强分类器的分类准确率．现实场景下的实验表明,本文算法在采用室内外环境中的WiFi信号、全球卫星导航系统（GNSS）可用星数、光照强度这三种融合的泛在信号进行室内外场景感知时性能最佳,对于不同角度方向下的室内外场景切换,可以在误报率仅为1.7％的情况下,达到98％的识别准确率,验证了本文所提算法的准确性和鲁棒性． 

Robust perception algorithm for indoor and outdoor scenes based on signal of opportunity

Considering that the development of navigation application service requirements for the combination of indoor and outdoor scenes and the problems of low recognition stability and the low recognition accuracy of existing indoor and outdoor scenes perception methods, this paper proposes a robust sensing algorithm for indoor and outdoor scenes based on signal of opportunity. The signal of opportunity is used to reduce the single signal recognition error. In order to improve the classification accuracy of the traditional AdaBoost algorithm for imbalanced data sets, the Probabilistic Neural Network(PNN) is used as the training weak classifier, and the entropy weight method is introduced to modify the weight of the weak classifier generated by iteration to improve the classification accuracy of the strong classifier. Experimental verification in real scenarios shows that the algorithm in this paper performs best in indoor and outdoor scene perception using signal of opportunity: WiFi signal, available GNSS stars, and light intensity in indoor and outdoor environments. For scenes switching in different angle directions, the recognition accuracy of 98％ can be achieved with a false alarm rate of only 1.7％, which verifies the accuracy and robustness of the algorithm proposed in this paper.