Indoor-Outdoor Scene Recognition Method based on Adaboost-PNN

Zhixian Cheng, Yongyi Mao, Chengkai Tian

Abstract: In response to the challenges of low accuracy and instability in indoor-outdoor seamless positioning technology for scene recognition, this paper proposes a high-precision indoor-outdoor scene recognition method based on an adaptive boosting algorithm (Adaboost-PNN). The method utilizes the built-in magnetic sensor, Global Navigation Satellite System (GNSS) module, and WiFi module of a smartphone to collect training data. Based on the different features exhibited by the number of satellites, signal-to-noise ratio, geomagnetic intensity, and WiFi signal strength in indoor and outdoor environments, the data features are input into the adaptive boosting algorithm to train the indoor-outdoor scene recognition model. Comprehensive testing results demonstrate that the indoor-outdoor scene recognition model based on the adaptive boosting algorithm achieves a recognition accuracy of 96.3% in different scenes, effectively distinguishing between indoor and outdoor environments. Moreover, compared to traditional recognition algorithms such as SVM, KNN, and PNN, the proposed method shows an improvement of 2%-5% in accuracy.