Masafumi Hashimoto, Mitsuo Yuminaka, and Kazuhiko Takahashi
People tracking, Ground laser scanner, Interacting multiple model estimator, Automatic calibration
In this paper, we present a people tracking system using static sensor nodes set at different locations in an environment. Our system has two sensor nodes equipped with a four-layer laser scanner and a central server. When the sensor nodes are allocated, a communication network between the sensor nodes and server is automatically constructed. The laser distance measurements are captured by the sensor nodes and sent to the central server. Based on the distance measurements, the server estimates the relative pose of two sensor nodes by ccorresponding vector fitting sample and consensus (CVFSAC) method before it begins to detect and track people. The server extracts the distance measurements of people by applying the background subtraction method. By using the distance measurements of people, heuristic-rule-based and global-nearest-neighbor (GNN)-based data association identifies people in a crowded environment. The interacting-multi-model (IMM) estimator tracks people with various motions such as walking, running or stopping. Experimental results based on tracking 18 people show the performance of the laser-based people tracking system.