Fusion of evidential occupancy grids for cooperative perception

Abstract

In the field of autonomous navigation, perception of the driving scene is one of the essential elements. Existing solutions combine on-board exteroceptive sensors, and are capable of understanding some of the near vehicle's dynamic surrounding environment. Furthermore, the perception capability of each vehicle can be enhanced by wireless information sharing if vehicles in the neighborhood transmit pertinent information. The primary benefit of such approach is to enable and improve the performance-safety of cooperative autonomous driving. The introduction of such a vehicle-to-vehicle communication leads to think of the global architecture as a system of systems. In this work, we address the task of evidential occupancy grid fusion so that a given vehicle can refine and complete its occupancy grid with the help of grids received from other near vehicles. The communication channel is supposed to be ideal, noiseless with infinite capacity. We focus on the fusion framework itself, using the theory of belief functions for reasoning with uncertainties on the relative poses of the vehicles and on the exchanged sensor measurement data. We evaluate the fusion system with real data acquired on public roads, with two connected vehicles.

A Review of Research on Traffic Conflicts Based on Intelligent Vehicles

“…Sensors can identify and track targets, but the reality will be blocked by other vehicles and infrastructure, resulting in perception information reduction. Camarda et al [94] used the fusion framework of lidar and wireless communication technology to realize the urban scene representation. The uncertainties of vehicle attitude and the measured data of switched sensors are used to improve the vehicle perception.…”

Multi-Agent Cooperative Camera-Based Evidential Occupancy Grid Generation

“…In the latter, the back-projected images on the ground are merged by using the union method. However, our merging method is based on Dempster Shaffer Theory [21], [22] as presented in [9]. We have also used the average occupancy fusion method based on the intersection between detections to assess the efficiency of each fusion method in our vision-based cooperative context.…”

“…2) Evidential merging method: We can also use the Dempster-Shafer Theory (DST) [21] to merge the LOGs as attempted in [9]. We give two possible statuses forming the universe given by the equation (7).…”

“…In this work, we take advantage of the in-scene point of view from the vehicles in order to enrich a common map of the scene in the form of an evidential occupancy grid [9] and to build confidence while decreasing the costs. Usually, classical occupancy grids as defined in [10] are used [11], [12].…”

Cooperative Raw Sensor Data Fusion for Ground Truth Generation in Autonomous Driving

“…Typically, more information is shared compared to object-level fusion requiring a higher data rate. In [14], occupancy maps of different vehicles are superimposed using Dempster-Shafer reasoning to yield a fused map. It has been further shown that erroneous estimations of vehicle states pose a major challenge in map alignment [15], [16].…”