Seed: A Segmentation-Based Egocentric 3D Point Cloud Descriptor for Loop Closure Detection

Fan, Yunfeng; He, Yichang; Tan, U-Xuan

doi:10.1109/iros45743.2020.9341517

Cited by 36 publications

(11 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…GOSMatch [22] employs only 3 kinds of semantic categories with 6 kinds of pairwise relations, and it uses histogram-based vertex and graph descriptors for matching. Seed [23] segments point clouds into objects and puts the polar BEV coordinate at a primary object before object projection. BoxGraph [24] uses the bounding box of objects as graph vertex and quantifies the pairwise similarity of vertices and edges to find optimal graph matching.…”

Section: Object and Graph Based Methodsmentioning

confidence: 99%

Fall Detection: a Camera and LiDAR Based Fall Detection System using Pose Estimation model (Preprint)

Wang¹,

Lai²

2022

Preprint

View full text Add to dashboard Cite

BACKGROUND Fall is dangerous for human, sometimes may cause body malfunction. Moreover, fall is dangerous especial for elder & newborn babies. According to statistics of World Health Organization, falls are the second leading cause of unintentional injury deaths worldwide. Each year an estimated 684,000 individuals die from falls globally of which over 80% are in low and middle income countries. Adults older than 60 years of age suffer the greatest number of fatal falls. 37.3 million of falls that are severe enough that require medical attention each year. It is possible to minimize the damage if we can make medical treatment as soon as possible after fall. We propose a new mechanism to detect fall using camera and LiDAR technology. OBJECTIVE Our goal is to make fall detection as precise as possible. We take advantage of camera and LiDAR to get the world 3D coordinate of the scene with human, and through pose estimation model to get human skeleton. Then we can do fall analysis using the 3D coordinate of human body. We can save more time for rescuers to rescue someone after fall. The scenario application of this system can be home, hospital, road, construction site, even factory, any place with human being. METHODS We take a photo of human using a camera and LiDAR. Then we use the 2D camera photo as the input of pose estimation machine learning model to get the 2D coordinate of human body. Then we transfer these 2D coordinates of human body into 3D coordinates by the 2D coordinates data and the depth information of LiDAR. Then we can detect if a man/woman is falling or not according to the 3D coordinate of the human body. RESULTS We analyze 4 poses of 5 rounds of 3 testers. The accuracy of this system is good (92.5%) even when parts of the body go out of camera sight. Both male and female and child testers with various height could be correctly detected (male of 182 cm, female of 160 cm and child of 98 cm). A fall can be detected and alerted in real time (more than 20 frames per second). Although there are some false negative when testing a fall pose. But many frames are analyzed during the single fall. Most of fall correctly alerted is enough for a rescue. CONCLUSIONS 3D coordinate is more precise and easier to do judgement than 2D coordinate. We can get 3D coordinate by combining the 2D coordinate of camera and the depth information of LiDAR. And Apple already launched some iOS and iPadOS device with camera and LiDAR. In this paper we use iPad Pro 11” 3rd edition to do this work, which means this mechanism is of low cost and easy to implement.

show abstract

Section: Object and Graph Based Methodsmentioning

confidence: 99%

Fall Detection: a Camera and LiDAR Based Fall Detection System using Pose Estimation model (Preprint)

Wang¹,

Lai²

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Readers are referred to [1] for a more thorough review on general loop closure methods. Existing research on 3D-based loop detection can be categorised into three groups [22]: feature-based [7], [8], [11], [23]- [25], segmentation-based [26], [27], and learningbased methods [9], [10], [28], [29].…”

Section: Related Workmentioning

confidence: 99%

“…eigenvalue-based and shape histograms, to describe each segment and performs the matching with random forest with RANSAC-based geometric verification. Seed [27] proposes handcrafted features that encode the topological information of segmented objects to reduce the noise and resolution effect. Both SegMatch and Seed uses the cluster-all method to segment the point cloud and requires the ground plane removal prior to the segmentation.…”

Section: Related Workmentioning

confidence: 99%

Loop closure detection using local 3D deep descriptors

Zhou¹,

Wang²,

Poiesi³

et al. 2021

Preprint

View full text Add to dashboard Cite

We present a simple yet effective method to address loop closure detection in simultaneous localisation and mapping using local 3D deep descriptors (L3Ds). L3Ds are emerging compact representations of patches extracted from point clouds that are learned from data using a deep learning algorithm. We propose a novel overlap measure for loop detection by computing the metric error between points that correspond to mutually-nearest-neighbour descriptors after registering the loop candidate point cloud by its estimated relative pose. This novel approach enables us to accurately detect loops and estimate six degrees-of-freedom poses in the case of small overlaps. We compare our L3D-based loop closure approach with recent approaches on LiDAR data and achieve state-of-theart loop closure detection accuracy. Additionally, we embed our loop closure approach in RESLAM, a recent edge-based SLAM system, and perform the evaluation on real-world RGBD-TUM and synthetic ICL datasets. Our approach enables RESLAM to achieve a better localisation accuracy compared to its original loop closure strategy.

show abstract

“…3D points above the ground plane level are converted to egocentric polar coordinates and projected to a 2D plane. This idea was extended in a couple of later works [4,8].…”

Section: Related Workmentioning

confidence: 99%

Large-Scale Topological Radar Localization Using Learned Descriptors

Komorowski¹,

Wysoczanska²,

Trzciński³

2021

Preprint

View full text Add to dashboard Cite

In this work, we propose a method for large-scale topological localization based on radar scan images using learned descriptors. We present a simple yet efficient deep network architecture to compute a rotationally invariant discriminative global descriptor from a radar scan image. The performance and generalization ability of the proposed method is experimentally evaluated on two large scale driving datasets: MulRan and Oxford Radar RobotCar. Additionally, we present a comparative evaluation of radar-based and LiDAR-based localization using learned global descriptors. Our code and trained models are publicly available on the project website. 1

show abstract

Seed: A Segmentation-Based Egocentric 3D Point Cloud Descriptor for Loop Closure Detection

Cited by 36 publications

References 20 publications

Fall Detection: a Camera and LiDAR Based Fall Detection System using Pose Estimation model (Preprint)

Fall Detection: a Camera and LiDAR Based Fall Detection System using Pose Estimation model (Preprint)

Loop closure detection using local 3D deep descriptors

Large-Scale Topological Radar Localization Using Learned Descriptors

Contact Info

Product

Resources

About