01658920

Component

Data collection techniques and methods to classify and track multiple objects are utilized in a wide range of traffic and safety studies. While new sensing technologies and data processing methods are being introduced for researchers and engineers, the metrics and methods for evaluating the performance of these new methods have received limited attention in the literature. Most of the evaluation studies to date relied on some form of benchmark method intended to provide the, so-called, ground truth. None of the methods is perfect, but it should be expected that the benchmark method will be considerably more accurate than the evaluated method. With the introduction of LiDAR methodology, it is becoming increasingly difficult to find a benchmark method that is both applicable to reasonably large-scale 3D measurements and considerably better than the LiDAR-based method. This paper aims to fill this knowledge gap by presenting a hybrid method of video-based tracking that, although it involves a human observer, is efficient enough to make the method practical. The key element of the proposed method is restitution of the 2D objects’ trajectories estimated on the video image plane with human assistance into the trajectories in the real-world 3D coordinates. The method was tested and then applied to evaluate a selected LiDAR-based tracking system and was found suitable for the task. This paper presents the main results of the exercise, the identified limitations of the evaluation method, and its possible improvements.

This paper was sponsored by TRB committee ABJ50 Standing Committee on Information Systems and Technology.

18-04250

English

14p

2018

Transportation Research Board 97th Annual Meeting

Digital/other

Figures; Photos; References; Tables

Algorithms; Detection and identification systems; Evaluation and assessment; Image processing; Laser radar; Pedestrians; Vehicle trajectories; Video

Data and Information Technology; Highways; Pedestrians and Bicyclists; Planning and Forecasting

Transportation Research Board Annual Meeting 2018 Paper #18-04250

TRIS, TRB, ATRI

Jan 8 2018 11:02AM