A recognition framework based on common cause detection in multimodal signals

为了实现真实的环境中的人和物体识别，本文提出了一种基于人的感知方式的多传感器集成方法。先前的传感器集成方法在特征提取和抽象之后组合不同种类的传感器（“识别后集成”）。我们提出了一种新的方法，结合从不同类型的传感器在早期阶段的感觉信号，并将其应用于人的识别和跟踪问题。通过使用激光测距仪（LRFs）跟踪环境中行人的腿部，同时观察行人的加速度信号。我们关联这些信号来自同一行人的基础上提出的信号相关方法，侧重于当地步行节奏。我们在普适和嵌入式计算与通信系统国际会议（PECCS 2012，全文接受率17%）上发表了这项研究，并获得了最佳论文奖。

To realize people and object recognition in real environment, we have proposed new multi sensor integration approach in the manner of human perception. Previous approaches of sensor integration combined different kinds of sensors after feature extraction and abstraction("integration after recognition"). We have proposed a new approach that combines sensory signals from different kinds of sensors in the earlier stage, and applied it to the problem of people identification and tracking. The legs of pedestrians in the environment are tracked by using laser range finders(LRFs), and acceleration signals from pedestrians are simultaneously observed. We associate these signals from same pedestrian based on the proposed signal correlation method that focuses on local waling rhythm. We has presented this study at the Int. Conf. on Pervasive and Embedded Computing and Communication Systems(PECCS 2012, full paper acceptance ratio 17%) and obtained the best paper award.