Traffic flow velocity estimation by virtually calibrated monocular surveillance cameras

Estimating traffic flow velocity is an important task in traffic condition assessment providing essential information for intelligent transportation systems (ITS). In order to measure vehicle speed, conventional methods utilize data extracted from multiple cameras where the times and locations of considering vehicles can be easily tracked. However, in practices there are a large number of residential and household security cameras deployed separately on the streets which can be utilized for traffic information extraction. In addition, there are no predefined landmarks under the sense of the considering camera to identify the moving distance of a vehicle throughout time frames making traffic flow velocity estimation from monocular cameras challenging. This paper proposes novel approaches to resolving these issues by utilizing reference objects which are commonly available on the sense. We propose mechanisms to estimate the pixel-per-meter ratio (PPMR) by examining the predefined size of the reference object. This ratio is used to estimate moving distances of vehicles within a time period which are then used to calculate vehicles’ velocity. In addition, essential difficulties including identifying and tracking reference objects, dealing with the change of PPMR in accordance with object’s positions on the image, and the changes of camera angles have been thoroughly resolved. Several experiments using real-world datasets including images and videos from the traffic surveillance camera systems managed by the Department of Traffic and Transport (DoTT), Ho Chi Minh City (HCMC), Vietnam have been conducted to validate the effectiveness and efficiency of the proposed method. The experimental results reveal that the proposed method is effective with the error rate of around 7.4% on the BrnoCompspeed dataset and 3% on the datasets collected in HCMC. This result reveals that the proposed approach is relevant to be extended for real-world applications, especially in the developing countries where traffic infrastructures have not been well developed and crowd-sourced commodity cameras are the main data sources.