JET-Net: Real-Time Object Detection for Mobile Robots

In most applications for autonomous robots, the detection of objects in their environment is of significant importance. As many robots are equipped with cameras, this task is often solved by image processing techniques. However, due to limited computational resources on mobile systems, it is common to use specialized algorithms that are highly adapted to the respective scenario. Sophisticated approaches such as Deep Neural Networks, which recently demonstrated a high performance in many object detection tasks, are often difficult to apply. In this paper, we present JET-Net (Just Enough Time), a model frame for efficient object detection based on Convolutional Neural Networks. JET-Net is able to perform real-time robot detection on a NAO V5 robot in a robot football environment. Experiments show that this system is able to reliably detect other robots in various situations. Moreover, we present a technique that reuses the learned features to obtain more information about the detected objects. Since the additional information can entirely be learned from simulation data, it is called Simulation Transfer Learning.

[1]  Thomas Röfer,et al.  SimRobot - A General Physical Robot Simulator and Its Application in RoboCup , 2005, RoboCup.

[2]  Manuela M. Veloso,et al.  SSL-Vision: The Shared Vision System for the RoboCup Small Size League , 2009, RoboCup.

[3]  Manuela M. Veloso,et al.  Visual sonar: fast obstacle avoidance using monocular vision , 2003, Proceedings 2003 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2003) (Cat. No.03CH37453).

[4]  Kaiming He,et al.  Feature Pyramid Networks for Object Detection , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[5]  Arne Hasselbring,et al.  A JIT Compiler for Neural Network Inference , 2019, RoboCup.

[6]  Nicolás Cruz,et al.  Using Convolutional Neural Networks in Robots with Limited Computational Resources: Detecting NAO Robots while Playing Soccer , 2017, RoboCup.

[7]  Ali Farhadi,et al.  You Only Look Once: Unified, Real-Time Object Detection , 2015, 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[8]  Alexander Fabisch,et al.  Robot Recognition and Modeling in the RoboCup Standard Platform League , 2010 .

[9]  Ali Farhadi,et al.  XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks , 2016, ECCV.

[10]  Ali Farhadi,et al.  YOLO9000: Better, Faster, Stronger , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[11]  Bo Chen,et al.  MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications , 2017, ArXiv.

[12]  Ian Goodfellow,et al.  Generative adversarial networks , 2020, Commun. ACM.

[13]  Sven Behnke,et al.  Learning Visual Obstacle Detection Using Color Histogram Features , 2011, RoboCup.

[14]  Wei Liu,et al.  SSD: Single Shot MultiBox Detector , 2015, ECCV.

[15]  Peter Stone,et al.  Fast and Precise Black and White Ball Detection for RoboCup Soccer , 2017, RoboCup.

[16]  Timo Aila,et al.  Pruning Convolutional Neural Networks for Resource Efficient Inference , 2016, ICLR.

[17]  Tim Laue,et al.  Vision-Based Orientation Detection of Humanoid Soccer Robots , 2017, RoboCup.

[18]  Stefan Wermter,et al.  Ball Localization for Robocup Soccer Using Convolutional Neural Networks , 2016, RoboCup.

[19]  Marcus M. Scheunemann,et al.  Deep Learning for Semantic Segmentation on Minimal Hardware , 2018, RoboCup.

[20]  Ali Farhadi,et al.  YOLOv3: An Incremental Improvement , 2018, ArXiv.

[21]  Ross B. Girshick,et al.  Focal Loss for Dense Object Detection , 2017, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[22]  Kaiming He,et al.  Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks , 2015, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[23]  Forrest N. Iandola,et al.  SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <1MB model size , 2016, ArXiv.

[24]  Jan Hoffmann,et al.  A Vision Based System for Goal-Directed Obstacle Avoidance , 2004, RoboCup.

[25]  Nicolás Cruz,et al.  Playing Soccer without Colors in the SPL: A Convolutional Neural Network Approach , 2018, RoboCup.

[26]  Norman Hendrich,et al.  ImageTagger: An Open Source Online Platform for Collaborative Image Labeling , 2018, RoboCup.