01627754

Component

This research proposes to estimate the column boundaries of the processing region, skip frames and update reference frames to improve vehicle detection and tracking. By estimating the column boundaries of the vehicle detection and tracking region, the authors track vehicles in almost the whole input image rather than vehicles in the overlapping area between the initial image and the subsequent image. The boundaries of the processing region are calculated based on the relative position of the reference frame and the non-reference frame only. The experimental result shows that the number of tracked vehicles decreases without estimating the column boundaries of the vehicle detection and tracking region, while it remains steady and only fluctuates slightly with estimating the column boundaries of the vehicle detection and tracking region. In the baseline, no frames are skipped in the first cycle after the initial road mask is estimated. However, some frames are skipped in subsequent cycles, in order to reduce the computation time. The experimental result shows that skipping every other frame is the best solution, i.e. the number of tracked vehicles and number of tracked non-vehicle blobs are close, and the total computation time is reduced, compared to inc=1 (inc represents the increment of the frames.). When the frame number increases, the overlap area between the reference frame (the first frame) and the subsequent frames becomes smaller and smaller. Hence, the quality of selected ground control points becomes worse and worse. Hence, considering the registration quality, updating reference frames is proposed. The experimental result shows that when updating the reference frame, the number of tracked vehicles is close compared to not updating the reference frame. However, the number of tracked non-vehicle blobs is significantly less in most frames, compared to not updating the reference frame.

This paper was sponsored by TRB committee AFB80 Standing Committee on Geospatial Data Acquisition Technologies.

01618707

17-04184

English

20p

2017

Transportation Research Board 96th Annual Meeting

Digital/other

Figures; References; Tables

Aerial reconnaissance; Automatic vehicle detection and identification systems; Image processing

Data and Information Technology; Highways; Vehicles and Equipment

Transportation Research Board Annual Meeting 2017 Paper #17-04184

TRIS, TRB, ATRI

Dec 8 2016 11:36AM