self-driving cars 'learn' to predict pedestrian movement - laser light measurement

Scientists use the human gait, the symmetry of the body and the position of the foot to teach themselves.
Driving a car recognizes and predicts pedestrian movements with higher accuracy than current technology.
Scientists use the human gait, the symmetry of the body and the position of the foot to teach themselves.
Driving a car recognizes and predicts pedestrian movements with higher accuracy than current technology.
Data collected by vehicles through cameras, lidar and GPS (GPS)
Allows researchers at the University of Michigan in the United States to capture video clips of human movements and then recreate them with three clipsdimensional (3D)
Computer simulation
Thus, they created a "network of circular neural networks inspired by bioengineering" to classify human movements.
The network can help predict the posture and future position of one or several pedestrians, about 50 yards from the vehicle, about the size of the city's intersections.
Lidar is a measurement method to measure the distance to the target by pulse laser irradiation of the target and using the sensor to measure the reflection pulse.
"Previous work in this area usually only looks at still images.
Ram Vasudevan, an assistant professor at the University of Michigan, said: "It doesn't really care how people move in three dimensions . ".
"But if these vehicles are going to run and interact in the real world, we need to make sure that our predictions of where pedestrians are going are not consistent with what the vehicle is going to do next," vasudwan said.
The necessary predictive power to equip vehicles requires a network to gain insight into the details of human movements: the speed of human gait (periodicity)
, Mirror symmetry of the limb, and the way the feet are placed to affect stability during walking.
Many machine learning that takes autonomous technology to the current level has processed two-dimensional images ---still photos.
A computer that shows millions of photos of parking signs will eventually identify parking signs in real-time in the real world.
However, by utilizing video clips running for a few seconds, the system can study the first half of the clips to make predictions, and then use the second half to verify accuracy.
"Right now, we're training the system to recognize the movement, not only to make predictions about one thing ---
Whether it's a parking sign or not--
But where will the body of the pedestrian be next, next and next, "said Matthew Johnson --
Robertson, associate professor at the University of Michigan.
"If a pedestrian is playing with their phone, you know they will be distracted," Vasudevan said . ".
"Their posture and how they look tell you their level of attention.
"It also tells you a lot about what they can do next," he said . ".
The results show that this new system improves the ability of driverless cars to identify what is most likely to happen next.