Deep Object Tracking on Dynamic Occupancy Grid Maps Using RNNs

Abstract

The comprehensive representation and understanding of the driving environment is crucial to improve the safety and reliability of autonomous vehicles. In this paper, we present a new approach to establish an environment model containing a segmentation between static and dynamic background and parametric modeled objects with shape, position and orientation. Multiple laser scanners are fused into a dynamic occupancy grid map resulting in a 360 • perception of the environment. A single-stage deep convolutional neural network is combined with a recurrent neural network, which takes a time series of the occupancy grid map as input and tracks cell states and its corresponding object hypotheses. The labels for training are created unsupervised with an automatic label generation algorithm.The proposed methods are evaluated in real-world experiments in complex inner city scenarios using the aforementioned 360 • laser perception. The results show a better object detection accuracy in comparison with our old approach as well as an AUC score of 0.946 for the dynamic and static segmentation. Furthermore, we gain an improved detection for occluded objects and a more consistent size estimation due to the usage of time series as input and the memory about previous states introduced by the recurrent neural network.

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

“…Authors in [10] and [4] employ Dynamic OGMs (DOGMa). DOGMa is the result of fusing a variety of sensor readings using Bayesian filtering, which associates dynamic information to each cell as well as the occupancy state [18].…”

“…Alternatively, it is possible to provide multi-steps of OGM prediction into the future [2,4,10]. Multi-step prediction of OGMs in a dynamic environment provides the drivable space for a planning algorithm without the need for the several stages required in the classic approach.…”

“…Recurrent Neural Networks (RNNs) are, therefore, a potential model candidate, which have also been successfully applied to various other problems with dynamic nature, such as unsupervised video prediction [12,7], multi-step prediction of mobile robots [15] and visual odometry [25]. Moreover, it is shown that RNNs can produce promising results in the context of OGM prediction [2,4].…”

Dynamic Environment Prediction in Urban Scenes using Recurrent Representation Learning

“…In contrast to continuous approaches, occupancy grids can generate a probabilistic map without having knowledge of the entire environment or making assumptions regarding the agent behavior by directly incorporating raw sensor measurements [9]. Furthermore, occupancy grids facilitate the use of common perception techniques such as object detection [10] and tracking [11]. Bayesian methods are often used to incorporate sensor measurements into occupancy grids [8].…”

Long-Term Occupancy Grid Prediction Using Recurrent Neural Networks

“…Deep learning approaches, applied on grid maps have showed promising results due to the image-like data structure, which allows to rely on standard convolutional neural network architectures [11]. Dynamic occupancy grid maps are utilized in [12] and [13] to address the task of object detection; and in [14] to refine the separation of dynamic and static cells. Wirges et al [15] propose a multi-layer grid to encode several features of lidar point clouds for object detection.…”