Occupancy grid mapping for rover navigation based on semantic segmentation

Obstacle mapping is a fundamental building block of the autonomous navigation pipeline of many robotic platforms such as planetary rovers. Nowadays, occupancy grid mapping is a widely used tool for obstacle perception. It foreseen the representation of the environment in evenly spaced cells, whose posterior probability of being occupied is updated based on range sensors measurement. In more classic approaches, the cells are updated to occupied at the point where the ray emitted by the range sensor encounters an obstacle, such as a wall. The main limitation of this kind of methods is that they are not able to identify planar obstacles, such as slippery, sandy, or rocky soils. In this work, we use the measurements of a stereo camera combined with a pixel labeling technique based on Convolution Neural Networks to identify the presence of rocky obstacles in planetary environment. Once identified, the obstacles are converted into a scan-like model. The estimation of the relative pose between successive frames is carried out using ORB-SLAM algorithm. The final step consists of updating the occupancy grid map using the Bayes’ update Rule. To evaluate the metrological performances of the proposed method images from the Martian analogous dataset, the ESA Katwijk Beach Planetary Rover Dataset have been used. The evaluation has been performed by comparing the generated occupancy map with a manually segmented ortomosaic map, obtained by drones’ survey of the area used as reference.

Semantic Segmentation With Labeling Uncertainty and Class Imbalance

Recently, methods based on Convolutional Neural Networks (CNN) achieved impressive success in semantic segmentation tasks. However, challenges such as the class imbalance and the uncertainty in the pixel-labeling process are not completely addressed. As such, we present a new approach that calculates a weight for each pixel considering its class and uncertainty during the labeling process. The pixel-wise weights are used during training to increase or decrease the importance of the pixels. Experimental results using different datasets like aerial, terrestrial, and ultrasound images show that the proposed approach leads to significant improvements in three challenging segmentation tasks in comparison to baseline methods. It was also proved to be more invariant to noise. The approach presented here may be used within a wide range of semantic segmentation methods to improve their robustness.

A Comparative Study of Two State-of-the-Art Feature Selection Algorithms for Texture-Based Pixel-Labeling Task of Ancient Documents

Recently, texture features have been widely used for historical document image analysis. However, few studies have focused exclusively on feature selection algorithms for historical document image analysis. Indeed, an important need has emerged to use a feature selection algorithm in data mining and machine learning tasks, since it helps to reduce the data dimensionality and to increase the algorithm performance such as a pixel classification algorithm. Therefore, in this paper we propose a comparative study of two conventional feature selection algorithms, genetic algorithm and ReliefF algorithm, using a classical pixel-labeling scheme based on analyzing and selecting texture features. The two assessed feature selection algorithms in this study have been applied on a training set of the HBR dataset in order to deduce the most selected texture features of each analyzed texture-based feature set. The evaluated feature sets in this study consist of numerous state-of-the-art texture features (Tamura, local binary patterns, gray-level run-length matrix, auto-correlation function, gray-level co-occurrence matrix, Gabor filters, Three-level Haar wavelet transform, three-level wavelet transform using 3-tap Daubechies filter and three-level wavelet transform using 4-tap Daubechies filter). In our experiments, a public corpus of historical document images provided in the context of the historical book recognition contest (HBR2013 dataset: PRImA, Salford, UK) has been used. Qualitative and numerical experiments are given in this study in order to provide a set of comprehensive guidelines on the strengths and the weaknesses of each assessed feature selection algorithm according to the used texture feature set.

FaceParser – A new face segmentation approach and labeleddatabase

Background and objective: A novel face parsing method is proposed in this paper which partition facial image into six semantic classes. Unlike previous approaches which segmented a facial image into three or four classes, we extended the class labels to six. Materials and Methods: A data-set of 464 images taken from FEI, MIT-CBCL, Pointing’04 and SiblingsDB databases was annotated. A discriminative model was trained by extracting features from squared patches. The built model was tested on two different semantic segmentation approaches – pixel-based and super-pixel-based semantic segmentation (PB_SS and SPB_SS).Results: A pixel labeling accuracy (PLA) of 94.68% and 90.35% was obtained with PB_SS and SPB_SS methods respectively on frontal images. Conclusions: A new method for face parts parsing was proposed which efficiently segmented a facial image into its constitute parts.