Reliable Attribute-Based Object Recognition Using High Predictive Value Classifiers
We consider the problem of object recognition in 3D using an ensemble of attribute-based classifiers. We propose two new concepts to improve classification in practical situations, and show their implementation in an approach implemented for recognition f
- PDF / 962,308 Bytes
- 15 Pages / 439.37 x 666.142 pts Page_size
- 31 Downloads / 223 Views
Institute for Systems Research, University of Maryland, College Park, USA {wluan,baras}@umd.edu 2 Computer Vision Lab, University of Maryland, College Park, USA [email protected], [email protected]
Abstract. We consider the problem of object recognition in 3D using an ensemble of attribute-based classifiers. We propose two new concepts to improve classification in practical situations, and show their implementation in an approach implemented for recognition from point-cloud data. First, the viewing conditions can have a strong influence on classification performance. We study the impact of the distance between the camera and the object and propose an approach to fusing multiple attribute classifiers, which incorporates distance into the decision making. Second, lack of representative training samples often makes it difficult to learn the optimal threshold value for best positive and negative detection rate. We address this issue, by setting in our attribute classifiers instead of just one threshold value, two threshold values to distinguish a positive, a negative and an uncertainty class, and we prove the theoretical correctness of this approach. Empirical studies demonstrate the effectiveness and feasibility of the proposed concepts.
1
Introduction
Reliable object recognition from 3D data is a fundamental task for active agents and a prerequisite for many cognitive robotic applications, such as assistive robotics or smart manufacturing. The viewing conditions, such as the distance of the sensor to the object, the illumination, and the viewing angle, have a strong influence on the accuracy of estimating simple as well as complex features, and thus on the accuracy of the classifiers. A common approach to tackle the problem of robust recognition is to employ attribute based classifiers, and combine the individual attribute estimates by fusing their information [1–3]. This work introduces two concepts to robustify the recognition by addressing common issues in the processing of 3D data, namely the problem of classifier dependence on viewing conditions, and the problem of insufficient training data. We first study the influence of distance between the camera and the object on the performance of attribute classifiers. Unlike 2D image processing techniques, which usually scale the image to address the impact of distance, depth based object recognition procedures using input from 3D cameras tend to be affected by distance-dependent noise, and this effect cannot easily be overcome [4]. c Springer International Publishing AG 2016 B. Leibe et al. (Eds.): ECCV 2016, Part III, LNCS 9907, pp. 801–815, 2016. DOI: 10.1007/978-3-319-46487-9 49
802
W. Luan et al.
We propose an approach that addresses effects of distance on object recognition. It considers the response of individual attribute classifiers’ depending on distance and incorporates it into the decision making. Though, the main factor studied here is distance, our mathematical approach is general, and can be applied to handle other factors affected by viewing conditions, such as light
Data Loading...
