SSF-PAN: Semantic Scene Flow-Based Perception for Autonomous Navigation in Traffic Scenarios
Yinqi Chen, Meiying Zhang, Qi Hao, Guang Zhou
AI summary
Problem
Traditional LiDAR navigation and SLAM systems suffer from high computational costs, accuracy degradation in dynamic traffic, and reliance on pre-built high-definition maps or separate moving-object detection modules.
Approach
The authors develop an iterative neural network that jointly estimates semantic scene flow and segments static versus dynamic objects using self-supervised geometric losses, integrated into a simulation-tested navigation platform.
Key results
- Reduces SLAM trajectory errors by over 93% compared to RANSAC baselines
- Achieves centimeter-level localization accuracy while maintaining real-time processing speeds
- Outperforms traditional moving object detection methods in navigation success rate and efficiency
- Delivers a robust, map-free navigation framework validated in CARLA simulations
Why it matters
Provides a computationally efficient, map-free perception pipeline that enhances the safety and scalability of autonomous vehicles in dynamic urban environments.
Abstract
Vehicle detection and localization in complex traffic scenarios pose significant challenges due to the interference of moving objects. Traditional methods often rely on outlier exclusions or semantic segmentations, which suffer from low computational efficiency and accuracy. The proposed SSF-PAN can achieve the functionalities of LiDAR point cloud based object detection/localization and SLAM (Simultaneous Localization and Mapping) with high computational efficiency and accuracy, enabling map-free navigation frameworks. The novelty of this work is threefold: 1) developing a neural network which can achieve segmentation among static and dynamic objects within the scene flows with different motion features, that is, semantic scene flow (SSF); 2) developing an iterative framework which can further optimize the quality of input scene flows and output segmentation results; 3) developing a scene flow-based navigation platform which can test the performance of the SSF perception system in the simulation environment. The proposed SSF-PAN method is validated using the SUScape-CARLA 3 and the KITTI [1] datasets, as well as on the CARLA simulator. Experimental results demonstrate that the proposed approach outperforms traditional methods in terms of scene flow computation accuracy, moving object detection accuracy, computational efficiency, and autonomous navigation effectiveness.