The article presents a method of using a super-resolution algorithm for a single camera recording road traffic. Vehicles smaller than a single pixel stimulate the entire pixel, making the estimation of their position and velocity difficult. The use of the Spatio-Temporal Track-Before-Detect (ST-TBD) algorithm allows for tracking sub-pixel objects, however, there are trajectories for which sub-pixel estimation is not possible. The proposed solution is based on a camera that is slightly shifted vertically and horizontally to record the scene image, allowing for the super-resolution effect to be achieved for data fusion in the state space. The article analyzes a 1D tracking configuration corresponding to the worst possible trajectory. For 40 different object velocities, the mean absolute position error was determined for a system with the ST-TBD algorithm and the same algorithm supported by an additional filter. The second solution, based on hierarchical tracking, allowed for more than a twofold reduction in the average maximum error and a threefold reduction in the average error.SUPER-RESOLUTION FROM KNOWN CAMERA MOTION FOR VEHICLES’ TRACKING USING SPATIO-TEMPORAL TRACK-BEFORE-DETECT ALGORITHM
Abstract
In the paper a method for application of the super-resolution algorithm for a single camera recording the road traffic is presented. Vehicles corresponding to the objects with size smaller than a single pixel stimulate the whole pixel, so the estimation of their location and velocity is troublesome. Using the Spatio-Temporal Track-Before-Detect (ST-TBD) algorithm it is possible to track the sub-pixel objects but there are some trajectories for which the sub-pixel estimation is not possible. Proposed solution is based on the slightly moving camera (horizontally and vertically) in order to record the images allowing to obtain the super-resolution effect and data fusion in the state space. In the paper the 1D tracking configuration has been analysed corresponding to the worst possible trajectory. The mean absolute error of location has been calculated for 40 different object’s velocities both for ST-TBD algorithm with and without additional filtering.
The second solution based on the hierarchical tracking allows for more than double reduction of the average maximum error and three times less average error. (…)
The article contains 19059 characters.
Source: Logistyka Magazine