Lightweight Active Object Retrieval with Weak Classifiers.

László Czúni,Metwally Rashad

doi:10.3390/s18030801

László Czúni, Metwally Rashad

Open Access

https://doi.org/10.3390/s18030801

Copy DOI

Journal: Sensors (Basel, Switzerland)	Publication Date: Mar 7, 2018
Citations: 5	License type: CC BY 4.0

Affiliation: University of Pannonia

Abstract

In the last few years, there has been a steadily growing interest in autonomous vehicles and robotic systems. While many of these agents are expected to have limited resources, these systems should be able to dynamically interact with other objects in their environment. We present an approach where lightweight sensory and processing techniques, requiring very limited memory and processing power, can be successfully applied to the task of object retrieval using sensors of different modalities. We use the Hough framework to fuse optical and orientation information of the different views of the objects. In the presented spatio-temporal perception technique, we apply active vision, where, based on the analysis of initial measurements, the direction of the next view is determined to increase the hit-rate of retrieval. The performance of the proposed methods is shown on three datasets loaded with heavy noise.

Highlights

The visual surveillance and recognition of 3D objects by unmanned aerial vehicles (UAVs) and by autonomous robots, and the replacement of bar/matrix codes with the objects’ views, are key technologies for future applications
There are two main topics that are closely related to our approach: video-based or multi-view object recognition and active vision, which is often found in robotics-related journals and conferences
The main contribution of this article, besides the analysis of the contribution of orientation information in the Hough framework, consists in showing that the proposed technique can be used for active perception and that very lightweight techniques can be efficiently used for 3D object retrieval

Summary

Introduction

The visual surveillance and recognition of 3D objects by unmanned aerial vehicles (UAVs) and by autonomous robots, and the replacement of bar/matrix codes (used for the identification of products) with the objects’ views, are key technologies for future applications. To achieve a high hit-rate, we continuously processed the images of the video camera while the object was targeted, measured the orientation of the camera, and evaluated the possible candidates with the help of compact visual descriptors This approach can be implemented in a dynamic recognition model, where even the orientation of new image queries can lead to faster and more accurate recognition. There has been a significant evolution in computer vision with the involvement of deep neural networks (DNNs), especially with convolutional neural networks (CNNs) for object detection and recognition These techniques can be considered as appearance-based approaches and can usually reach high recognition rates in many scenarios.