As one of today's popular research field, mobile robots, are widely used in entertainment, search and rescue, health, military, agriculture and many other fields with the advantages of technological developments. Object detection is one of the methods used for mobile robots to gather and report information about its environment during these tasks. With the ability to detect and classify objects, a robot can determine the type and number of objects around it and use this knowledge in its movement and path planning or reporting the objects with the desired features. Considering the dimensions of mobile robots and weight constraints of flying robots, the use of these algorithms is more limited. While the size and weight of mobile devices should be kept relatively small, successful object classification algorithms require processors with high computational power. In this study, to be able to use object detection information for mapping and path planning, object detection and classification methods were examined, and for the usage in low weight and low energy consuming platforms through developer boards, detection algorithms were compared to each other.
Nesne tespit ve sınıflandırma yöntemlerinin başarısının artırılması amacıyla model odaklı ve veri odaklı yaklaşımlar araştırmacılar tarafından son yıllarda sıklıkla çalışılmaktadır. Araştırmacıların birçoğu problemlere özgü model önerilerinde bulunmakta ve mevcut modeller üzerinde değişimler önermektedir. Öte yandan, eğitim sürecinde kullanılmakta olan veri üzerinde yapılan çalışmaların sayışa oldukça azdır. Bu çalışmada, mevcut bir tanıma ve sınıflandırma problemi üzerinde, model ve veri odaklı yaklaşımların etkileri kıyaslanmıştır. Yaygın kullanıma sahip olan YOLOv4 ağı üzerinde yapılan ağ yapısı değişikliğinin başarım ve performansa etkisiyle, veri setinde kullanılan verilerin yeniden hazırlanmasıyla elde edilen başarım karşılaştırılarak yorumlanmıştır. Ağ yapısının değişimi ile nesne tanıma başarısı yaklaşık %4 oranında artarken, hesaplama hızında ortalama %8’lik düşüş meydana gelmiştir. Öte yandan verilerin yeniden hazırlanarak nesne tanıma algoritmasının çalıştırılması %6 oranında kazanç sağlarken, hesaplama maliyetinde değişime neden olmamıştır. Günümüzde yeteri kadar dikkate alınmasa da veri üzerindeki hazırlıkların sınıflandırma doğruluğuna önemli derecede etki yaptığı gözlemlenmiştir.
Unknown closed spaces are a big challenge for the navigation of robots since there are no global and pre-defined positioning options in the area. One of the simplest and most efficient algorithms, the artificial potential field algorithm (APF), may provide real-time navigation in those places but fall into local minimum in some cases. To overcome this problem and to present alternative escape routes for a robot, possible crossing points in buildings may be detected by using object detection and included in the path planning algorithm. This study utilized a proposed sensor fusion method and an improved object classification method for detecting windows, doors, and stairs in buildings and these objects were classified as valid or invalid for the path planning algorithm. The performance of the approach was evaluated in a simulated environment with a quadrotor that was equipped with camera and laser imaging detection and ranging (LIDAR) sensors to navigate through an unknown closed space and reach a desired goal point. Inclusion of crossing points allows the robot to escape from areas where it is congested. The navigation of the robot has been tested in different scenarios based on the proposed path planning algorithm and compared with other improved APF methods. The results showed that the improved APF methods and the methods reinforced with other path planning algorithms were similar in performance with the proposed method for the same goals in the same room. For the goals outside the current room, traditional APF methods were quite unsuccessful in reaching the goals. Even though improved methods were able to reach some outside targets, the proposed method gave approximately 17% better results than the most successful example in achieving targets outside the current room. The proposed method can also work in real-time to discover a building and navigate between rooms.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.