Non-cooperating vehicle tracking in VANETs using the conditional logit model

• Main Authors: Reza, Tahsin Arafat, Barbeau, Michel, Lamothe, Gilles, Alsubaihi, Badr
• Format: Conference Proceeding
• Language: English
• Subjects: Cameras; Roads; Target tracking; Trajectory; Vehicles; Vehicular ad hoc networks
• ISSN: 2153-0009; 2153-0017
• DOI: 10.1109/ITSC.2013.6728301

Vehicular Ad Hoc Networks (VANETs) are widely considered as indispensable elements of the future intelligent transportation systems that are aiming to apply information and communications technologies to improve transportation safety and quality of experience. We present our take on a relatively unexplored problem, exploiting VANETs for on-road surveillance. The proposal is inspired by multi-agent systems intended for surveillance, e.g., a distributed camera network. We propose a tracking system composed of three operational modules, namely, localization, tracking data collection and prediction of future locations of a target. Camera equipped onboard units (OBUs) act as remote mobile sensors. Tracking messages are communicated among the OBUs and roadside units (RSUs). These messages are also triggered in the possible locations of the target in a timely manner. Therefore, it is imperative to scope the search to limit the number of OBUs and RSUs involved in the tracking operation, thus, minimizing the number of tracking messages. To this end, a movement modeling technique utilizes the OBU-observations to classify the target's movement pattern to aid future trajectory prediction. In our previous work, we proposed a Dirichlet-multinomial (D-M) model under the Bayesian estimation framework. In this paper, we present newly identified cues towards improving the movement estimation model. The D-M model is constrained to the assumption that all the choice sets are identical across trials. We demonstrate that this is almost never the case. The improved model exploits a choice model, called the conditional logit. The conditional logit model is attractive when choice sets vary across trials. Additionally, we weight outcome of each trial according to the given choice sets to achieve higher estimation accuracy. We evaluate the new model by means of an experimental analysis and compare results with the D-M model.