Detailed environment perception is a crucial component of automated vehicles. However, to deal with the amount of perceived information, we also require segmentation strategies. Based on a grid map environment representation, well-suited for sensor fusion, free-space estimation and machine learning, we detect and classify objects using deep convolutional neural networks. As input for our networks we use a multi-layer grid map efficiently encoding 3D range sensor information. The inference output consists of a list of rotated bounding boxes with associated semantic classes. We conduct extensive ablation studies, highlight important design considerations when using grid maps and evaluate our models on the KITTI Bird's Eye View benchmark. Qualitative and quantitative benchmark results show that we achieve robust detection and state of the art accuracy solely using top-view grid maps from range sensor data.

There are several techniques for the performance of energy studies in buildings, where infrared thermography is widely used for the study of the composition of the envelopes, to find faults in the building materials and in the composition of the envelopes with influence on their thermal behaviour, as well as to detect areas with humidity. Common thermographic building interpretations are performed by a human operator, which involves a high level of subjectivity and mainly relies on the expertise of the operator. With the aim at minimizing this subjectivity and maximizing the accuracy of the inspections, this paper presents a procedure for the automation of thermographic building inspections mainly focused on thermal bridges. The procedure, in addition to detecting the thermal bridges by their geometric characteristics and their temperature differences with the surroundings, includes the computation of the thermophysical property of linear thermal transmittance of each candidate to thermal bridge, thus implying their characterization in addition to their detection. With this addition, together with a previous process of rectification of the thermal images analysed, the accuracy of the detection of thermal bridges regarding existing methodologies is improved in 15% considering the false positives and negatives obtained in each methodology.

Accessibility diagnosis of as-built urban environments is essential for path planning, especially in case of people with reduced mobility and it requires an in-depth knowledge of ground elements. In this paper, we present a new approach for automatically detect and classify urban ground elements from 3D point clouds. The methodology enables a high level of detail classification from the combination of geometric and topological information. The method starts by a planar segmentation followed by a refinement based on split and merge operations. Next, a feature analysis and a geometric decision tree are followed to classify regions in preliminary classes. Finally, adjacency is studied to verify and correct the preliminary classification based on a comparison with a topological graph library. The methodology is tested in four real complex case studies acquired with a Mobile Laser Scanner Device. In total, five classes are considered (roads, sidewalks, treads, risers and curbs). Results show a success rate of 97% in point classification, enough to analyse extensive urban areas from an accessibility point of view. The combination of topology and geometry improves a 10% to 20% the success rate obtained with only the use of geometry.

In the near future, the communication between autonomous cars will produce a network of sensors that will allow us to know the state of the roads in real time. Lidar technology, upon which most autonomous cars are based, allows the acquisition of 3D geometric information of the environment. The objective of this work is to use point clouds acquired by Mobile Laser Scanning (MLS) to segment the main elements of road environment (road surface, ditches, guardrails, fences, embankments, and borders) through the use of PointNet. Previously, the point cloud was automatically divided into sections in order for semantic segmentation to be scalable to different case studies, regardless of their shape or length. An overall accuracy of 92.5% has been obtained, but with large variations between classes. Elements with a greater number of points have been segmented more effectively than the other elements. In comparison with other point-by-point extraction and ANN-based classification techniques, the same success rates have been obtained for road surfaces and fences, and better results have been obtained for guardrails. Semantic segmentation with PointNet is suitable when segmenting the scene as a whole, however, if certain classes have more interest, there are other alternatives that do not need a high training cost.

Building accessibility diagnosis is of high interest especially in case of people with reduced mobility. This paper proposes a methodology for automated detection of inaccessible steps in building façade entrances from MLS (mobile laser scanner) data. Our approach uses the MLS trajectory to automatically subdivide urban point clouds into regular stretches. From each stretch, the lower zone of façade is isolated and selected as region of interest. Points belonging to vertical elements are projected onto a 2D image and steps are detected and classified as inaccessible areas according to the comparison of geometrical features such as height jump, proximity to ground and width, with regulation. The methodology has been tested in four real datasets, which constitute more than 400 meters of different urban scenarios. Results exhibit a robust performance under urban scenes with a high variability of façade geometry due to the presence of different entrance types to shops and dwellings. Results have been quantitatively evaluated and they show global F1 value around 93%. Moreover, the methodology is very fast since 100 m are processed in less than 2 minutes.

Scan planning of buildings under construction is a key issue for an efficient assessment of work progress. This work presents an automatic method aimed to determinate the optimal scan positions and the optimal route based on the use of Building Information Models (BIM) and considering data completeness as stopping criteria. The method is considered for a Terrestrial Laser Scanner mounted on a mobile robot following a stop & go procedure. The method starts by extracting floor plans from the BIM model according to the planned construction status, and including geometry and semantics of the building elements considered for construction control. The navigable space is defined from a binary map considering a security distance to building elements. After a grid-based and a triangulation-based distribution are implemented for generating scan position candidates, a visibility analysis is carried out to determine the optimal number and position of scans. The optimal route to visit all scan positions is addressed by using a probabilistic ant colony optimization algorithm. The method has been tested in simulated and real buildings under very dissimilar conditions and structural construction elements. The two approaches for generating scan position candidates are evaluated and results show the triangulation-based distribution as the more efficient approach in terms of processing and acquisition time, especially for large-scale buildings.

No abstract

Highlights: Direct use of point clouds to generate navigable graphs for pedestrians in urban environments  Real routes generated according to two motor skills: pedestrians without reduced mobility and wheelchairs  Navigable space based on ground elements and static objects  Occlusion correction of sidewalks by applying morphological operations