Brain4Cars (2016)

我们提出了一种车辆传感器丰富的平台和用于机动预期的学习算法。为此，我们为汽车配备了摄像头，全球定位系统 (GPS) 和计算设备，以捕获汽车内部和外部的驾驶环境。为了预测操作，我们提出了一种感觉融合深度学习体系结构，该体系结构共同学习预测和融合多个感觉流。我们的体系结构由递归神经网络 (rnn) 组成，该网络使用长短期记忆 (LSTM) 单元来捕获长时间依赖。我们提出了一种新颖的训练程序，该程序允许网络仅在部分时间背景下预测未来。我们介绍了1180英里自然高速公路和城市驾驶的各种数据集，并表明我们可以在实时发生之前3.5秒预测机动，并分别具有90.5% 和87.4% 的精度和召回率。

We propose a sensor-rich vehicle platform and a learning algorithm for maneuver anticipation. To this end, we equip vehicles with cameras, Global Positioning System (GPS) devices, and computing equipment to capture the driving environments inside and outside the car. For maneuver prediction, we propose a sensor-fusion deep learning architecture that jointly learns to predict and fuse multiple sensory streams. Our architecture consists of a Recurrent Neural Network (RNN) that uses Long Short-Term Memory (LSTM) units to capture long-term dependencies. We propose a novel training procedure that enables the network to predict the future using only partial temporal context. We introduce a diverse dataset of 1,180 miles of natural highway and urban driving, and demonstrate that we can anticipate maneuvers up to 3.5 seconds before they occur in real time, with precision and recall of 90.5% and 87.4% respectively.

搜集汇总

背景与挑战

背景概述

Brain4Cars (2016)是一个专注于车辆机动预测的自动驾驶数据集，包含1180英里的驾驶数据，采用深度学习体系结构实现高精度预测。数据集由多所知名大学和研究机构联合发布，适用于自动驾驶相关研究。