Publications
Summary
The purpose of this research, in partnership with SEPTA, is to address the problem of frequent bus-pedestrian accidents in dense urban areas through the creation of an intelligent pedestrian warning system. Our proposed system uses visual detection and parametric speaker technologies to provide a concise, focused beam of sound that reaches only a focused target while minimizing the amount of disruption and disturbance to neighboring areas. We believe that this solution will provide a minimally invasive, cost-effective method for reducing bus-pedestrian collisions and substantially reducing the loss of life and property.
Here are the major components:
- Bus Tracking: In order to detect the location of the bus, the system gathers raw data from onboard GPS and IMU sensors. This data is filtered in real time with an Unscented Kalman Filter and correlated with the expected path of the bus (from published timetables) in order to accurately anticipate when the bus is about to make a turn. Only during turns is the pedestrian collision detection active (at other times, we assume, it would produce too many spurious warnings from passing cars, bicycles, etc, and data shows that bus-pedestrian collisions are concentrated at intersections during turns). Much of this work was completed at Penn by Yida Zhang, Rahul Bhan, Nikhil Karnik, Jordan Parker, Vaibhav Wardhen, and Thomas Boutin.
- Collision Detection: Sensors mounted high up on the front corners of the bus scan for pedestrians who might be on a collision course. Currently, we use a Hokuyo scanning laser rangefinder with a 270 degree field of view. It scans 1000 times per second and gives very accurate distance readings. Unfortunately, it also costs an order of magnitude above our target price point for the entire system, so it will likely be replaced with a cheaper camera or sonar system, and a heuristic algorithm to make up for the loss in accuracy. The current basic system was contructed at Penn by me and Alaric Qin.
- Warning Delivery: This is where the majority of the recent work has been conducted, and where much of the novelty of our system is derived. Other warning systems use a siren or flashing lights to alert everyone within a certain radius of the turning bus to potential danger. But most of those people are not in danger, so it's disruptive and annoying. We use a directional loudspeaker to project sound only towards the detected pedestrian in danger. It's the opposite of the Cone of Silence. The technology is an array of ultrasonic speakers -- the ultrasonic waves travel in a narrow cone (instead of omnidirectionally like a normal speaker) and the nonlinear properties of the air itself convert the ultrasonic waves back to audible sound. The video below shows the camera moving around our prototype speaker, and the perceived volume changes drastically, even though the speaker is playing the same sound. This research was started at Penn by me, Alaric Qin, Marcus Pan and Justin Aird and is being continued by Alaric and Marcus.
Code
Demonstration
- Video showing the effectiveness of the directional speaker
