Environment Perception Framework Fusing Multi-Object Tracking, Dynamic Occupancy Grid Maps and Digital Maps
This work addresses the challenge of reliable perception for autonomous driving, though it appears incremental as it combines existing methods rather than introducing a fundamentally new approach.
The paper tackles the problem of robust environment perception for autonomous vehicles by fusing multi-object tracking and dynamic occupancy grid maps to compensate false positives, achieving improved robustness in both rural and urban scenarios as demonstrated with real-world data.
Autonomously driving vehicles require a complete and robust perception of the local environment. A main challenge is to perceive any other road users, where multi-object tracking or occupancy grid maps are commonly used. The presented approach combines both methods to compensate false positives and receive a complementary environment perception. Therefore, an environment perception framework is introduced that defines a common representation, extracts objects from a dynamic occupancy grid map and fuses them with tracks of a Labeled Multi-Bernoulli filter. Finally, a confidence value is developed, that validates object estimates using different constraints regarding physical possibilities, method specific characteristics and contextual information from a digital map. Experimental results with real world data highlight the robustness and significance of the presented fusing approach, utilizing the confidence value in rural and urban scenarios.