Inexpensive video cameras, radar, GPS, and other sensors are revolutionizing how transportation researchers analyze traffic. Collecting petabytes of highway data is easy. Processing and understanding it is not. The sheer volume of data is overwhelming the available analysis tools and is quickly becoming unmanageable.

Researchers from NREC cut through this bottleneck. Using machine learning and image processing techniques, our team automated the analysis of videos and other highway data. The system makes it easier and faster to study these large-scale data sets and extract usable information from them. These innovative methods give transportation researchers the keys to unlock the potential of the data they have collected.

The NREC approach combines traffic sign detection, vehicle detection, traffic signal light detection, and state estimation with advanced scene parsing and “chunking” to enable context driven scene understanding. The massive size of these data sets, the immense variety of techniques for collecting them, and the wide range of conditions under which they were recorded are all advantages for machine learning.

A software tool helps researchers to apply these analytical techniques to large sets of highway video data. Researchers can fine-tune the machine learning software by adjusting the input parameters for detecting vehicles, signs, traffic lights, the width of road shoulders, and other visual features of highways.

Automating the labor-intensive work of detecting, labeling, and measuring items of interest in the video data sets speeds up their analysis. This enables researchers to more quickly and easily identify highway design features that affect congestion, crashes, and other safety problems.