In autonomous driving, retrieving a specific traffic scene in huge datasets is significant challenge. Traditional retrieval methods struggle to cope with the semantic complexity and heterogeneity of scenes are unable meet variable needs different users. This paper proposes “Query-by-Example”, approach based on Visual-Large Language Model (VLM)-generated Road Scene Graph (RSG) representation. Our method uses VLMs generate structured graphs from video data, capturing high-level attributes...

Rich semantic information extraction plays a vital role on next-generation intelligent vehicles. Currently there is great amount of research focusing fundamental applications such as 6D pose detection, road scene segmentation, etc. And this provides us opportunity to think about how shall these data be organized and exploited. In paper we propose graph,a special scene-graph for Different classical representation, graph not only object proposals but also their pair-wise relationships. By...

Automated identification of the relationships between traffic actors and surrounding objects, in order to describe their behavior predict intentions, has become focus increasing attention field autonomous driving. Therefore, this work, we propose a Road Scene Graphs-Graph Convolutional Network (RSG-GCN) as novel, graph-based model for predicting topological graph structure given scene. The status HD map information are integrated prior knowledge, allowing edges linking actor nodes capture...

Currently, state-of-the-art simultaneous localization and mapping methods are capable of generating large-scale dense environmental maps. One primary reason may be the applications map partitioning strategies. An efficient method will decrease time complexity algorithm and, more importantly, make robots understand a place anthropomorphically. In this article, we propose novel segmentation based on quadtree spectral clustering. The is first organized hierarchically using quadtree, then...

Reproducing real-world traffic scenes in the simulator is fundamental to training self-driving systems. Creating a simulation scenario complex task, generally done manually: ego-vehicle and other entities are placed their trajectories defined, trying recreate some situation found real traffic. To reduce manual burden, here we propose Real-to-Synthetic toolset. This toolset provides synthetic scene openDrive format, which can be directly simulated many simulators such as SUMO or CARLA. Also,...

Behavioral and semantic relationships play a vital role on intelligent self-driving vehicles ADAS systems. Different from other research focused trajectory, position, bounding boxes, relationship data provides human understandable description of the object's behavior, it could describe an past future status in amazingly brief way. Therefore is fundamental method for tasks such as risk detection, environment understanding, decision making. In this paper, we propose RSG-Net (Road Scene Graph...

Map segmentation method similar to the way human percept external environment can decrease computational complexity of robot navigation algorithm and SLAM problem. This paper presents an introduction application spectral cluster in map process. Then this several kinds similarity measurement criteria construct matrix. With these criteria, mobile encountor different environments. Furthermore, a self-adaptive clustering based on silhouette coefficient criteria. As result that result, effective...

Browsing specific traffic scene in large-scale dataset is an increasing demand from researchers, self-driving community and insurance companies. It easy to search scenes with tags such as "rain", "snow", or "on highway". However, searching configurations, like "two vehicles waiting for a person crossing the road", still open problem. In this paper, we provide RSG-search, scene-graph based retrieval method, on our previous research generation. By previously translating datasets graphs, can...