Keywords
IEM-N (KL0-electronica), vehículo, kalman, colisión, predicción, seguridad, conducción, autónoma, vehicle, kalman, collision, prediction, security, driving, autonomous
Collision prediction/avoidance through connected smart cars
Abstract
Este Proyecto ha sido realizado por Roberto Rioja García en la Universisty of Texas at Austin bajo la supervisión del Dr. Joydeep Ghosh, el grupo de investigación CARSTOP y el Texas Department of Transportation (TxDOT). El objetivo de este proyecto fue el desarrollo y creación de un sistema de detección de colisiones y alerta para vehículos. Este sistema debe constituir una alternativa asequible para ayudar a reducir el número de accidentes de coches actual. El proyecto se enfoca en crear un sistema que funcione en tiempo real, que pueda predecir colisiones y alertar al conductor de manera exacta y con tiempo de margen para maniobrar. Esta únicamente diseñado para detectar posibles colisiones y alertar al conductor; nunca tomar acción para prevenir la colisión lo cual será responsabilidad del conductor. Los objetivos eran tener un detector de objetos con una exactitud superior al 46% obtenido por el proyecto de fin de grado previo, darle al conductor 3 segundos de margen de reacción antes de la colisión y mantener el ratio de falsos positivos por debajo de un 30%, todo ello procesando cada frame en menos de 250 milisegundos.
El sistema final está dividido en 8 submódulos completamente diferenciados e independientes entre sí: sensores, detector de objetos, sincronización LIDAR-cámara, transformación LIDAR-cámara, localizador de objetos, filtrado de Kalman, detector de colisiones y alerta.
En este proyecto se ha creado de manera exitosa un sistema de detección de colisiones con un alto nivel de exactitud. Se ha mejorado la exactitud a la hora de localizar objetos del equipo del año pasado con el sistema corriendo en menos de 250 ms. El error de falsos positivos sigue siendo superior al objetivo inicial, pero se ha reducido y se ha fallado a la hora de dar un tiempo de alerta superior a 3 segundos, el tiempo medio de reacción de un conductor.
This project was done by Roberto Rioja García at the University of Texas at Austin with the support of Dr. Ghosh, the CARSTOP research team and the Texas Department of Transportation. The project was focused on developing a vehicular collision detection and warning system. This system should constitute an affordable way to help cut down on the number of car crashes. The project was focused on creating a system that works in real time, can accurately predict collisions, and alert the driver in a timely fashion. It is only designed to detect possible collisions and alert the driver; this system does not take action to prevent these collisions from occurring. The goals were to have an object location accuracy higher than the 46% of the previous year´s team, to give the driver 3 seconds of warning for a collision, and to have a false positive rate below 30%, and process each frame in less than 250 milliseconds. The final system is divided in 8 distinct and independent subsystems: sensors, object detection, LIDAR-camera synchronization, LIDAR-camera transformation, object location, Kalman filtering, collision detection and alert. In this project, a system that detects collisions with a high level of accuracy was successfully created. Also achieving the goal of increasing our location accuracy over last year’s team and running time of less than 250 ms and came close to meeting the goal for false positives. However, it failed to meet the goal of providing a 3 second alert time, which is the average reaction time of a driver, because accuracy was considered a priority during the testing.
This project was done by Roberto Rioja García at the University of Texas at Austin with the support of Dr. Ghosh, the CARSTOP research team and the Texas Department of Transportation. The project was focused on developing a vehicular collision detection and warning system. This system should constitute an affordable way to help cut down on the number of car crashes. The project was focused on creating a system that works in real time, can accurately predict collisions, and alert the driver in a timely fashion. It is only designed to detect possible collisions and alert the driver; this system does not take action to prevent these collisions from occurring. The goals were to have an object location accuracy higher than the 46% of the previous year´s team, to give the driver 3 seconds of warning for a collision, and to have a false positive rate below 30%, and process each frame in less than 250 milliseconds. The final system is divided in 8 distinct and independent subsystems: sensors, object detection, LIDAR-camera synchronization, LIDAR-camera transformation, object location, Kalman filtering, collision detection and alert. In this project, a system that detects collisions with a high level of accuracy was successfully created. Also achieving the goal of increasing our location accuracy over last year’s team and running time of less than 250 ms and came close to meeting the goal for false positives. However, it failed to meet the goal of providing a 3 second alert time, which is the average reaction time of a driver, because accuracy was considered a priority during the testing.
Información detallada
Materias, derechos, colecciones e identificadores