Light Sensor Based Vehicle and Pedestrian Detection Method for Wireless Sensor Network

The paper proposes a method, which utilizes light sensors from wireless nodes, to detect moving objects like vehicles or pedestrians. The method is analyzing light intensity of the general red, green, and blue spectrums of visible light from nodes that are placed on a roadside. The proposed aggregation algorithm, based on justified granulation paradigm, adapts exponential forgetting mechanism to descriptive statistic functions (features). This approach allows to reduce memory utilization of wireless node. The aggregated values are used by lightweight state-of-the-art machine learning methods to build profile of moving objects. The method is tuned using heuristic-based genetic algorithm. Advantages of the introduced method were demonstrated in real-world scenarios. Broad experiments were conducted to test various classification approaches and feature subsets. The experimental results confirm that the introduced method can be adopted for sensor node, which can detect objects independently or in cooperation with other nodes (working as classifier ensemble).