N-body Filtering for Road Tracking using a Car Following Model Mahendra Mallick Propagation Research Associates, Inc. Marietta, GA 30066, USA mahendra.mallick@gmail.com Steve Rubin 2224 Pleasant Hill Lane Holiday, FL 34691, USA slr42349@gmail.com Jorge Laval Georgia Institute of Technology School of Civil and Environmental Engineering Atlanta, GA 30332, USA jorge.laval@ce.gatech.edu Abstract - Filtering for each track is performed independent of other tracks in conventional multi-target tracking algorithms. However, vehicles on roads follow inter-vehicle constraints and road constraints. Therefore, conventional tracking algorithms can produce many unrealistic results when applied to vehicles moving on roads. We present a new filtering algorithm, which enforces the inter-vehicle distance constraint to a group of N vehicles moving along a single lane of a road, using a car following model. Our filtering algorithm uses the interacting multiple model estimator to model the motion of maneuvering vehicles and enforces the inter-vehicle driving distance constraint among N vehicles. For testing and validating our algorithm, we use the US Highway 101 trajectory data, which is regarded as a benchmark data set for traffic flow theory. Keywords: Road tracking, Car following model (CFM), Traffic flow theory (TFT), Filtering with constraints, N- body filtering, Interacting multiple model estimator (IMM), Rauch-Tung-Streibel (RTS) smoother, US Highway 101 trajectory data. 1 Introduction In conventional multi-target tracking, filtering for each track is performed independent of other tracks. If conventional tracking algorithms are applied to vehicles moving on roads, then many unrealistic results can appear. The reason for this is that vehicles on roads follow road constraints (e.g. staying in lanes, speed limits, etc.) and inter-vehicle constraints (e.g. maintaining safe distances in a lane and lane change constraints) as shown in Figure 1. Conventional tracking is unconstrained in nature and ignores these constraints. However, the tracking algorithms have a number of desirable modeling features such as stochastic/statistical dynamic and measurement models [1-3], which are suitable for real-world applications. On the other hand, the traffic flow theory (TFT) [5-9], [15-16] considers the motion of groups of vehicles and uses road constraints and inter-vehicle constraints. However, the models are deterministic and the target state is non-random in nature. Secondly, measurements from sensors are not used to update the vehicle state in traffic flow theory. In this paper, we present a new filtering algorithm which enforces the inter-vehicle distance constraint to a group of N vehicles moving along a single lane of a road, using a car following model (CFM). Vehicle following model may be more appropriate. However, we use CFM, since this is commonly used in TFT. The motion of the vehicles includes nearly constant velocity (NCV) motion, accelerated and decelerated motions, and stopping. Therefore, our filtering algorithm uses the interacting multiple model (IMM) estimator [4], [1] to model the motion of a maneuvering vehicle and enforces the inter- vehicle driving distance constraint among N vehicles. Figure 1. Motion of vehicles on road following road constraints (e.g. staying in lanes, speed limits, etc.) and inter-vehicle constraints. For testing and validating our algorithms, we use the US Highway 101 trajectory data near Los Angeles area, which includes trajectories of 61 vehicles. US Highway 101 trajectory data is treated as the target truth trajectory. We generate simulated position measurements using 1000 Monte Carlo simulations. The paper is organized as follows. We describe the US Highway 101 trajectory data in Section 2. In Section 3, we present the dynamic and measurement models. Section 4 discusses details of our CFM. Estimation of slope and intercept parameters of our CFM using the US Highway 101 data is discussed in Section 5. Section 6 describes the N-body IMM filtering algorithm using our CFM. 14th International Conference on Information Fusion Chicago, Illinois, USA, July 5-8, 2011 978-0-9824438-3-5 ©2011 ISIF 102