Multi-Step Prediction of Occupancy Grid Maps With Recurrent Neural Networks

Abstract

We investigate the multi-step prediction of the drivable space, represented by Occupancy Grid Maps (OGMs), for autonomous vehicles. Our motivation is that accurate multi-step prediction of the drivable space can efficiently improve path planning and navigation resulting in safe, comfortable and optimum paths in autonomous driving. We train a variety of Recurrent Neural Network (RNN) based architectures on the OGM sequences from the KITTI dataset. The results demonstrate significant improvement of the prediction accuracy using our proposed difference learning method, incorporating motion related features, over the state of the art. We remove the egomotion from the OGM sequences by transforming them into a common frame. Although in the transformed sequences the KITTI dataset is heavily biased toward static objects, by learning the difference between consecutive OGMs, our proposed method provides accurate prediction over both the static and moving objects. A video of the performance of our method on the KITTI dataset is available at https: //youtu.be/Bskd0Z7eLFE.

Algorithm for Complexing Multiple Data Sources Into a Single Occupancy Map

“…Для детекции динамических объектов в карте занятости, как правило, используются изображения c камер и нейросетевые подходы [5,6]. Другое семейство алгоритмов для обнаружения динамики также использует сверточные сети, но в качестве входа подаются классические карты занятости [7,8]. В то же время конкурсы по обнаружению и трекингу трехмерных движущихся объектов [9,10] показывают, что подавляющее большинство наиболее точных методов используют данные лидара и выдают информацию в виде вектора с координатами, ориентацией и размерами объекта [11,12].…”

CF-LSTM: Cascaded Feature-Based Long Short-Term Networks for Predicting Pedestrian Trajectory

“…A traditional approach having considered both static and dynamic interactions is a named "Social Force" model (Helbing and Molnar 1995), defining interactions as forces upon the pedestrians. Modern socially-aware methods usually use recurrent neural networks (Mikolov et al 2010) for trajectory prediction (Fernando et al 2018a;Kitani et al 2011;Ryoo et al 2014;Srivastava, Mansimov, and Salakhudinov 2015;Vemula, Muelling, and Oh 2018;Liang et al 2019;Chung et al 2014;Mohajerin and Rohani 2019;Liu et al 2016;Shi et al 2019) and introduce attention mechanism to interaction measure (Bhattacharyya, Fritz, and Schiele 2018;Choi and Dariush 2019) and social behavior understanding (Sadeghian et al 2019;Haddad et al 2019;Al-Molegi, Jabreel, and Martínez-Ballesté 2018;Varshneya and Srinivasaraghavan 2017). Also, Generative Adversarial Network (GAN) model is designed to generate multiple reasonable trajectories (Gupta et al 2018;Fernando et al 2018b;Amirian, Hayet, and Pettré 2019).…”

Visionnet: A Coarse-To-Fine Motion Prediction Algorithm Based On Active Interaction-Aware Drivable Space Learning

“…Secondly, there are many factors that are relevant to the trajectory of an agent, such as the nature of the ground, the mental state of the driver and the pedestrian, etc. To address these challenges, many trajectory prediction methods have been proposed which can be roughly classified into two categories, i.e., the coordinate-based methods [1,2,3,4] and the vision-based methods [5,6,7,8,9].…”

“…However, the interaction features obtained by these coordinate-based methods do not have any physical meaning that can be visualized for interpretation. On the other hand, the vision-based methods are able to use semantic images, including camera images, LiDAR point cloud data, and Occupancy Grid Maps (OGMs) [5], for global interaction feature extraction. Work [6] generate the prediction sequences in the form of images by Convolutional Neural Networks (CNNs).…”