Xiangyi Qin

Xiangyi Qin

Auto-generated profile, based on publicly available information

Xiangyi Qin

School of Intelligent Systems Engineering, Shenzhen Sun Yat-Sen University, Shenzhen, China

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
19 November 2025 in Information Fusion
Verified
Journal Article
Closed access
XZxiaofei zhang
xiaofei zhang
YLYining Li
Yining Li
JWJinping Wang
Jinping Wang
XQXiangyi Qin
Xiangyi Qin
Ying ShenYing Shen
Ying Shen
Xiaojun TanXiaojun Tan
Xiaojun Tan

28 June 2024 in Measurement Science and Technology
Verified
Journal Article
Closed access
Xiaojun TanXiaojun Tan
Xiaojun Tan
XQXiangyi Qin
Xiangyi Qin
NCNanjie Chen
Nanjie Chen
LZLetian Zhang
Letian Zhang
LJLu Jie
Lu Jie
Linlin YouLinlin You
Linlin You
LiDAR odometry is a critical part of LiDAR simultaneous localization and mapping (SLAM). However, existing methods often suffer from the gradual accumulation of errors. Furthermore, the intensive computational requirements of processing large-scale global landmarks make it impractical to directly introduce bundle adjustment(BA) into real-time odometry. To overcome these challenges, this article presents a new strategy named precise landmark-map for BA odometry. First, BA odometry is optimized by a new active landmark maintenance strategy, aiming to improve local registration accuracy and mitigate error accumulation. Specifically, in contrast to conventional methods that only retain feature points within the sliding window, this paper retains all stable landmarks on the map and removes landmarks based on their level of activity. Moreover, computational efficiency is improved by minimizing the sliding window size and implementing marginalization to maintain scans that are outside the window but associated with active landmarks on the map. In addition, experiments on three challenging datasets validate the real-time performance of our algorithm in outdoor driving scenarios, outperforming state-of-the-art LiDAR SLAM algorithms like Lego-LOAM and VLOM.

31 July 2024 in arXiv
Verified
Preprint
Open access
PDF file
XZxiaofei zhang
xiaofei zhang
YLYining Li
Yining Li
JWJinping Wang
Jinping Wang
XQXiangyi Qin
Xiangyi Qin
Ying ShenYing Shen
Ying Shen
Xiaojun TanXiaojun Tan
Xiaojun Tan
Perception systems of autonomous vehicles are susceptible to occlusion, especially when examined from a vehicle-centric perspective. Such occlusion can lead to overlooked object detections, e.g., larger vehicles such as trucks or buses may create blind spots where cyclists or pedestrians could be obscured, accentuating the safety concerns associated with such perception system limitations. To mitigate these challenges, the vehicle-to-everything (V2X) paradigm suggests employing an infrastructure-side perception system (IPS) to complement autonomous vehicles with a broader perceptual scope. Nevertheless, the scarcity of real-world 3D infrastructure-side datasets constrains the advancement of V2X technologies. To bridge these gaps, this paper introduces a new 3D infrastructure-side collaborative perception dataset, abbreviated as inscope. Notably, InScope is the first dataset dedicated to addressing occlusion challenges by strategically deploying multiple-position Light Detection and Ranging (LiDAR) systems on the infrastructure side. Specifically, InScope encapsulates a 20-day capture duration with 303 tracking trajectories and 187,787 3D bounding boxes annotated by experts. Through analysis of benchmarks, four different benchmarks are presented for open traffic scenarios, including collaborative 3D object detection, multisource data fusion, data domain transfer, and 3D multiobject tracking tasks. Additionally, a new metric is designed to quantify the impact of occlusion, facilitating the evaluation of detection degradation ratios among various algorithms. The Experimental findings showcase the enhanced performance of leveraging InScope to assist in detecting and tracking 3D multiobjects in real-world scenarios, particularly in tracking obscured, small, and distant objects. The dataset and benchmarks are available at https://github.com/xf-zh/InScope.

3 June 2026 in IEEE Transactions on Multimedia
Verified
Journal Article
Closed access
XQXiangyi Qin
Xiangyi Qin
XZxiaofei zhang
xiaofei zhang
SWShuai Wang
Shuai Wang
YWYuzhen Wei
Yuzhen Wei
JWJinping Wang
Jinping Wang
Xiaojun TanXiaojun Tan
Xiaojun Tan
Advancements in vision-language models (VLMs) have generated interest in language-based driving approaches. However, existing methods are limited to single-step visual question answering (VQA), whereas human drivers use multimodal information to synthesize a coherent chain of thought (CoT). To close this gap, we propose Cognidrive, a VLM-based cognitive driving understanding framework. In Cognidrive, a Cognidrive-Agent is used to interact with real-world driving scenes, a Reasoning module is applied to extract object information, and a Knowledge module is used to record environmental knowledge. This framework enables VLMs to perform multistep environmental interactions, generating human-interpretable multimodal CoT reasoning for driving decision-making. Specifically, the Cognidrive-Agent uses a multiview image projector to extract features from surrounding-view images, and incorporates object spatial positional embeddings to encode the spatial locations of traffic objects. Moreover, we propose the M$^{2}$Drive instruction dataset to evaluate and enhance multistep multimodal CoT reasoning capabilities of VLMs. The dataset consists of 58 k QA pairs for 8 subtasks, featuring diverse reasoning materials and extended reasoning chains. Experiments with the M$^{2}$Drive and DriveLM datasets demonstrate that Cognidrive significantly outperforms five strong baseline models and achieves state-of-the-art performance.