Probabilistic Fusion of Vehicle Features for Reidentification and Travel Time Estimation Using Video Image Data

Abstract

This paper proposes a probabilistic vehicle reidentification algorithm for estimating travel time using the image data provided by traffic surveillance cameras. Each vehicle is characterized by its color, type, and length, which are extracted from the video record using image processing techniques. A data fusion rule is introduced to combine these three features to generate a probabilistic measure for a reidentification (matching) decision. The vehicle-matching problem is then reformulated as a combinatorial problem and solved by a minimum-weight bipartite matching method. To reduce the computational time, the algorithm uses the potential availability of historic travel time data to define a potential time window for vehicle reidentification. This probabilistic approach does not require vehicle sequential information and hence allows vehicle reidentification across multiple lanes. The algorithm is tested on a 5-km section of the expressway system in Bangkok, Thailand. The travel time estimation result is also compared with the directly observed data.