東華大學圖書館 |

語系: 繁體中文

說明(常見問題)

回圖書館首頁

手機版館藏查詢

登入

回首頁 到查詢結果 [ null ]

切換: 標籤 | MARC模式 | ISBD

Dynamic Obstacle Detection and Motio...

Colino, Francisco Jose Barbosa Marques,

FindBook

Google Book

Amazon

博客來

Dynamic Obstacle Detection and Motion Prediction for an AMR Using Stereo-Vision Based Fusion /

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Dynamic Obstacle Detection and Motion Prediction for an AMR Using Stereo-Vision Based Fusion // Francisco Jose Barbosa Marques Colino.
作者:	Colino, Francisco Jose Barbosa Marques,
面頁冊數:	1 electronic resource (196 pages)
附註:	Source: Masters Abstracts International, Volume: 86-06.
Contained By:	Masters Abstracts International86-06.
標題:	Robots. -
電子資源:	https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31787762
ISBN:	9798346841203

Dynamic Obstacle Detection and Motion Prediction for an AMR Using Stereo-Vision Based Fusion /
Colino, Francisco Jose Barbosa Marques,

Dynamic Obstacle Detection and Motion Prediction for an AMR Using Stereo-Vision Based Fusion /Francisco Jose Barbosa Marques Colino. - 1 electronic resource (196 pages)

Source: Masters Abstracts International, Volume: 86-06.

In dynamic and unpredictable environments like industrial warehouses, Autonomous Mobile Robots (AMRs) face challenges in detecting and tracking static and dynamic obstacles. Existing stereo vision solutions often struggle with range limitations, detection accuracy, and handling occlusions, hindering practical use. Additionally, reliance on expensive sensors like LiDAR or purely deep learning-based methods is not always feasible in cost-sensitive or variable environments. To address these challenges, this dissertation proposes a cost-effective, real-time stereo vision-based fusion framework for dynamic obstacle detection and motion prediction.The main objectives were to develop a perception system using stereo cameras for depth sensing, improve obstacle detection through hybrid techniques, especially for human agents, and implement an advanced tracking system that handles occlusions and non-linear object movement. Additionally, the system aims to generate a dynamic occupancy grid integrating future movement predictions, supporting path planning and ensuring modularity for future use.The framework consists of several key modules. Ermis uses DBSCAN for unsupervised 3D clustering, offering flexibility in detecting objects without prior training, useful in environments with unknown obstacles. Periklis, using the YOLOv8 model, complements this by recognizing object classes in 2D and projecting them into 3D space. The fusion of these methods enhances detection accuracy and adaptability, allowing detection of both known and unknown objects. Ganelis integrates outputs from both methods, reducing false positives and ensuring a comprehensive scene understanding. Adonis, with an enhanced OC-SORT algorithm and Growth-Area Association, tracks objects through occlusions, predicting their movements and ensuring continuous tracking, improving system robustness. The dynamic occupancy grid, incorporating both current and future obstacle positions, facilitates integration with path planning and enables real-time trajectory adjustments, improving safety and efficiency.The system was evaluated using custom simulations in NVIDIA Isaac Sim, testing various conditions including static, dynamic, and occlusion scenarios. Performance was measured using F1-score, average translation error, and scale error. The results showed significant improvements in detection accuracy and tracking reliability, especially in handling occlusions and long-distance detection. The system's compatibility with edge AI platforms, such as NVIDIA Jetson, demonstrates its potential as a cost-effective and energy-efficient alternative to LiDAR-based solutions.

English

ISBN: 9798346841203Subjects--Topical Terms:

529507
Robots.

Dynamic Obstacle Detection and Motion Prediction for an AMR Using Stereo-Vision Based Fusion /
LDR:03985nmm a22003853i 4500 001 2432558
005 20250718090829.5
006 m o d
007 cr|nu||||||||
008 261011s2024 miu||||||m |||||||eng d
020 $a 9798346841203
035 $a (MiAaPQD)AAI31787762
035 $a (MiAaPQD)10216162418
035 $a AAI31787762
040 $a MiAaPQD $b eng $c MiAaPQD $e rda
100 1 $a Colino, Francisco Jose Barbosa Marques, $e author. $3 3821836
245 1 0 $a Dynamic Obstacle Detection and Motion Prediction for an AMR Using Stereo-Vision Based Fusion / $c Francisco Jose Barbosa Marques Colino.
264 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2024
300 $a 1 electronic resource (196 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Masters Abstracts International, Volume: 86-06.
500 $a Advisors: Pinto, Vitor Hugo; Ribeiro, Francisco.
502 $b M.C.E. $c Universidade do Porto (Portugal) $d 2024.
520 $a In dynamic and unpredictable environments like industrial warehouses, Autonomous Mobile Robots (AMRs) face challenges in detecting and tracking static and dynamic obstacles. Existing stereo vision solutions often struggle with range limitations, detection accuracy, and handling occlusions, hindering practical use. Additionally, reliance on expensive sensors like LiDAR or purely deep learning-based methods is not always feasible in cost-sensitive or variable environments. To address these challenges, this dissertation proposes a cost-effective, real-time stereo vision-based fusion framework for dynamic obstacle detection and motion prediction.The main objectives were to develop a perception system using stereo cameras for depth sensing, improve obstacle detection through hybrid techniques, especially for human agents, and implement an advanced tracking system that handles occlusions and non-linear object movement. Additionally, the system aims to generate a dynamic occupancy grid integrating future movement predictions, supporting path planning and ensuring modularity for future use.The framework consists of several key modules. Ermis uses DBSCAN for unsupervised 3D clustering, offering flexibility in detecting objects without prior training, useful in environments with unknown obstacles. Periklis, using the YOLOv8 model, complements this by recognizing object classes in 2D and projecting them into 3D space. The fusion of these methods enhances detection accuracy and adaptability, allowing detection of both known and unknown objects. Ganelis integrates outputs from both methods, reducing false positives and ensuring a comprehensive scene understanding. Adonis, with an enhanced OC-SORT algorithm and Growth-Area Association, tracks objects through occlusions, predicting their movements and ensuring continuous tracking, improving system robustness. The dynamic occupancy grid, incorporating both current and future obstacle positions, facilitates integration with path planning and enables real-time trajectory adjustments, improving safety and efficiency.The system was evaluated using custom simulations in NVIDIA Isaac Sim, testing various conditions including static, dynamic, and occlusion scenarios. Performance was measured using F1-score, average translation error, and scale error. The results showed significant improvements in detection accuracy and tracking reliability, especially in handling occlusions and long-distance detection. The system's compatibility with edge AI platforms, such as NVIDIA Jetson, demonstrates its potential as a cost-effective and energy-efficient alternative to LiDAR-based solutions.
546 $a English
590 $a School code: 5896
650 4 $a Robots. $3 529507
650 4 $a Data processing. $3 680224
650 4 $a Visualization. $3 586179
650 4 $a Robotics. $3 519753
650 4 $a Statistics. $3 517247
690 $a 0771
690 $a 0463
710 2 $a Universidade do Porto (Portugal). $e degree granting institution. $3 3815379
720 1 $a Pinto, Vitor Hugo $e degree supervisor.
720 1 $a Ribeiro, Francisco $e degree supervisor.
773 0 $t Masters Abstracts International $g 86-06.
790 $a 5896
791 $a M.C.E.
792 $a 2024
856 4 0 $u https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31787762