SwRI’s human motion prediction system overlays sparse skeletal features on pedestrians to predict movement. Credit: Southwest Research Institute.

Southwest Research Institute (SwRI) in the US has developed a motion prediction system designed to strengthen the pedestrian detection capabilities of autonomous vehicles (AVs).

The new system enables AVs to identify sudden changes in pedestrian movements, unlike existing technologies that predict the linear movements of pedestrians and other obstacles.

The motion prediction system observes real-time biomechanical movements around the pedestrian’s pelvic area to predict sudden movement changes.

SwRI senior research analyst Samuel E Slocum said: “For instance, if a pedestrian is walking west, the system can predict if that person will suddenly turn south. As the push for automated vehicles accelerates, this research offers several important safety features to help protect pedestrians.”

Typically, motion prediction technology uses optical flow, a form of computer vision that pairs algorithms with cameras to track dynamic objects.

To develop the new technology, SwRI compared this optical flow to other deep learning methods such as temporal convolutional networks (TCNs) and long short-term memory (LSTM) to identify an optimised configuration.

The research also utilised SwRI’s markerless motion capture system, which automates biomechanical analysis.

The optimised system uses a convolutional neural network to process video data and predict abrupt movements in milliseconds, with higher levels of accuracy.

SwRI Applied Sensing Department manager Dr Douglas Brooks said: “If we see a pedestrian, we might prepare to slow down or change lanes in anticipation of someone crossing the street.

“We take it for granted, but it’s incredibly complex for a computer to process this scene and predict scenarios.”