Robust Place Recognition Using Illumination-compensated Image-based Deep Convolutional Autoencoder Features
- PDF / 2,691,572 Bytes
- 9 Pages / 594.77 x 793.026 pts Page_size
- 4 Downloads / 227 Views
ISSN:1598-6446 eISSN:2005-4092 http://www.springer.com/12555
Robust Place Recognition Using Illumination-compensated Image-based Deep Convolutional Autoencoder Features Chansoo Park, Hee-Won Chae, and Jae-Bok Song* Abstract: Place recognition is a method for determining whether a robot has previously visited the place it currently observes, thus helping the robot correct its accumulated position error. Ultimately, the robot will travel long distances more accurately. Conventional image-based place recognition uses features extracted from a bag-of-visualwords (BoVW) scheme or pre-trained deep neural network. However, the BoVW scheme does not cope well with environmental changes, and the pre-trained deep neural network is disadvantageous in that its computation time is high. Therefore, this paper proposes a novel place recognition scheme using an illumination-compensated imagebased deep convolutional autoencoder (ICCAE) feature. Instead of reconstructing the raw image, the autoencoder designed to extract ICCAE features is trained to reconstruct the image, whose illumination component is compensated in the logarithm frequency domain. As a result, we can extract the ICCAE features based on a convolution layer that is robust to illumination and environmental changes. Additionally, ICCAE features can perform faster feature matching than the features extracted from existing deep networks. To evaluate the performance of ICCAE feature-based place recognition, experiments were conducted using a public dataset that includes various conditions. Keywords: Convolutional autoencoder, frequency image, illumination compensation, place recognition.
1.
INTRODUCTION
Recently, most simultaneous localization and mapping (SLAM) systems use a loop-closure detection algorithm that recognizes a location previously visited by a robot to correct its accumulated positional error [1]. To perform robust loop-closure detection, camera-based place recognition methods have been actively studied in recent years. The well-known fast appearance-based mapping (FAB-MAP) [2] and real-time appearance-based mapping (RTAB-MAP) [3] are the place recognition techniques that use the bag-of-visual-words (BoVW) method [4]. Additionally, ORB-SLAM2 [5], a famous SLAM system, demonstrated stable performance by performing place recognition using a bag of binary words for a fast place recognition method (DBoW2) [6]. BoVW-based techniques classify hand-crafted visual features into word vectors and use them for place recognition. The BoVW-based method is fast. However, it exhibits the disadvantage of increasing false-positive probability when various environmental changes occur. For example, Fig. 1 shows that a place may be misrecognized because of dynamic obstacles and variations in appearance. To compensate for this,
Fig. 1. Comparison of images from the same location at different times: (a) image at daytime, and (b) image at night. SeqSLAM [7], which uses sequence information of an image, has been proposed. However, this method is vulnerable to large changes in
Data Loading...
