A real-time dynamic obstacle tracking and mapping system for UAV navigation and collision avoidance with an RGB-D camera (2209.08258v4)

Abstract: The real-time dynamic environment perception has become vital for autonomous robots in crowded spaces. Although the popular voxel-based mapping methods can efficiently represent 3D obstacles with arbitrarily complex shapes, they can hardly distinguish between static and dynamic obstacles, leading to the limited performance of obstacle avoidance. While plenty of sophisticated learning-based dynamic obstacle detection algorithms exist in autonomous driving, the quadcopter's limited computation resources cannot achieve real-time performance using those approaches. To address these issues, we propose a real-time dynamic obstacle tracking and mapping system for quadcopter obstacle avoidance using an RGB-D camera. The proposed system first utilizes a depth image with an occupancy voxel map to generate potential dynamic obstacle regions as proposals. With the obstacle region proposals, the Kalman filter and our continuity filter are applied to track each dynamic obstacle. Finally, the environment-aware trajectory prediction method is proposed based on the Markov chain using the states of tracked dynamic obstacles. We implemented the proposed system with our custom quadcopter and navigation planner. The simulation and physical experiments show that our methods can successfully track and represent obstacles in dynamic environments in real-time and safely avoid obstacles. Our software is available on GitHub as an open-source ROS package.

• The paper presents a hybrid 3D mapping approach that fuses occupancy voxel maps with depth-image region proposals for precise obstacle detection.
• The paper introduces a dual-filter method combining Kalman and continuity filters to reliably identify and track dynamic obstacles.
• The paper demonstrates effective real-time trajectory prediction using a Markov chain method, achieving over 25Hz performance on UAV hardware.

Overview of Real-Time Dynamic Obstacle Tracking and Mapping for UAVs

This paper introduces a sophisticated system designed for real-time dynamic obstacle tracking and mapping to aid Unmanned Aerial Vehicles (UAVs) in navigating and avoiding collisions. The authors focus on utilizing RGB-D cameras to provide the quadcopters with enhanced perception capabilities, addressing constraints typically faced by UAVs, such as limited computational resources.

The proposed solution leverages a 3D hybrid map system that utilizes an occupancy voxel map for efficient static environment representation, alongside employing depth images to generate obstacle region proposals. This system effectively distinguishes between static and dynamic obstacles—a significant limitation in conventional voxel-based methods. Furthermore, it introduces a strategy for real-time obstacle tracking using a combination of Kalman and continuity filters, providing robust performance even with restricted onboard computations.

Key Contributions

• Region Proposal Detector with Map Refinement: By employing a lightweight depth image-based detector, the system generates region proposals for potential obstacles that are immediately refined using static map data. This hybrid method ensures improved estimation of obstacle positioning and sizing.
• Dynamic Obstacle Identification and Tracking: The system enables continuous tracking of dynamic obstacles by implementing both the Kalman filter for velocity estimation and a novel continuity filter to ensure accurate dynamic classification, reducing false positives from static objects.
• Environment-Aware Trajectory Prediction: Utilizing a Markov chain-based method, the trajectory prediction module incorporates environment interaction, enhancing prediction accuracy concerning obstacle movement relative to static map data.

Simulation experiments showcase the system's efficiency in handling environments with static structures and multiple moving entities, demonstrating accurate real-time obstacle detection and trajectory prediction. Physical experiments further reinforce its capability to sustain performance when implemented on actual UAV systems in various test scenarios.

Experimental Findings

Quantitative analysis places the proposed system at an advantage compared to state-of-the-art methods, specifically in terms of dynamic obstacle position and velocity estimation. The track record of the UAV system using this approach reflected a reduced failure ratio in trajectory predictions, particularly in complex environments. Performance metrics affirm the method's capability to sustain over 25Hz operational frequency on an Nvidia Xavier NX, validating its suitability for real-time deployment.

Implications and Future Directions

The practical implications of implementing a robust system such as this are significant, allowing UAVs to operate effectively in dense or unpredictable environments typical in applications like urban navigation or search and rescue. The system's ability to utilize low-computation solutions means its application can extend to a range of UAV models, enhancing scalability and adaptability.

The authors note the potential for exploring camera models with broader fields of view to further improve tracking accuracy in dynamic environments. This could unlock new dimensions in UAV autonomy, allowing for more seamless interaction within varied operational spaces.

In conclusion, this paper presents a cogent and methodological advancement in UAV dynamic obstacle mapping and tracking, which could serve as a foundation for enhancing autonomy in future aerial robotics applications. The balance between computational efficiency and robust dynamic environment interaction marks a noteworthy step forward in UAV capabilities.