Néstor Morales

Real-time adaptive obstacle detection based on an image database

Abstract In this paper, an innovative method for the detection and avoidance of obstacles is presented. This is based on image registration techniques. The aim of this method is the detection of the possible obstacles that could be in the route where VERDINO (an autonomous electrical vehicle which is going to work in the surroundings of a bioclimatic urbanization as part of the SAGENIA project) navigates. The obstacle detection is one of the most critical parts of the prototype. It is responsible for the detection and later avoidance of the pedestrians, cars, etc. that can damage it or be damaged by the prototype. The algorithm is able to work in real time, with good detection rates and a fast response. It also includes a dynamical database that will allow the vehicle to learn adaptively the current state of the environment, rejecting old images. With this, some of the typical problems related to the image registration techniques are removed. Some examples and results are provided, corroborating the good behavior of the algorithm, which has been tested in simulated and real conditions. The method is also applicable to other tasks, like surveillance for non-stationery cameras or other applications over very different kinds of images.

Generating automatic road network definition files for unstructured areas using a multiclass support vector machine

Smooth and safe paths.Tolerance to the sensor measurement errors.First method in using Multiclass SVM for path planning.Better results in smoothness than other related SVM path planning methods.Similar results in terms of safety than other related SVM path planning methods. In this paper, an innovative methodology for the generation of a Road Network Definition File (RNDF) using only an obstacle map as input is presented. This RNDF, which relies on a Multiclass Support Vector Machine(MSVM)-based trajectory generation method, will be used by an autonomous vehicle for transporting people in closed, unstructured areas for which no previous information is available, such as residential areas or industrial parks. The advantages of using this technique are the generation of a safe and smooth trajectory graph (making the trip more comfortable for riders by having trajectories pass as far away as possible from obstacles). Moreover, although there exist other previous Support Vector Machine (SVM) path planning methods, this is the first to use a MSVM. The advantages of doing so are that by obtaining a decision boundary for each object in the scene, all possible trajectories are computed and joined to form a graph. This is done through a combination of a Nearest-Neighbor Graph (NNG) and a Relative Neighborhood Graph (RNG). The method was tested with real data and in real conditions, yielding good results. At the end of the paper, results for two kinds of studies are presented. The first set of tests is intended to determine the best parameter values for the proposed methodology. In the second set of evaluations, the approach is compared with other state-of-the-art SVM-based methods, as well as with a classical approach, demonstrating that the method outperforms them in some aspects. Furthermore, the source code of the method is available for testing, as are some videos in which the output of the method is shown, including a comparison with previous methods.

Path planning using a Multiclass Support Vector Machine

Graphical abstractDisplay Omitted HighlightsGeneration of a smooth path based on the hyperplane generated by the SVM training stage.The margin maximization constraint of the SVM ensures that the generated path will be as far as possible from the obstacles in the environment.Robustness against uncertainty and the noise generated by sensors.Generation of a decision graph that allows using higher level methods to select the final path.The method is compared with a previous approach and in real world conditions. In this paper, a new path planning algorithm for unstructured environments based on a Multiclass Support Vector Machine (MSVM) is presented. Our method uses as its input an aerial image or an unfiltered auto-generated map of the area in which the robot will be moving. Given this, the algorithm is able to generate a graph showing all of the safe paths that a robot can follow. To do so, our algorithm takes advantage of the training stage of a MSVM, making it possible to obtain the set of paths that maximize the distance to the obstacles while minimizing the effect of measurement errors, yielding paths even when the input data are not sufficiently clear. The method also ensures that it is able to find a path, if it exists, and it is fully adaptable to map changes over time. The functionality of these features was assessed using tests, divided into simulated results and real-world tests. For the latter, four different scenarios were evaluated involving 500 tests each. From these tests, we concluded that the method presented is able to perform the tasks for which it was designed.

Object detection in non-stationary video surveillance for an autonomous vehicle

In this paper, a new method for the automated video surveillance of wide areas is described. Using the images obtained from a set of cameras installed on an autonomous vehicle, a video surveillance tool has been developed, based on the comparison between images that have been taken in the same place but at different times. The vehicle drives around the watched area, looking for intruders. The method described in this paper is the image comparison system used for this task, and it is based on image registration and change detection techniques. The system has been fully tested, obtaining promising results. The validation process shows the good performance of the methods selected to develop the application. It is also able to be executed in real time with good detection rates.

Safe and reliable navigation in crowded unstructured pedestrian areas

In this paper, the navigation system of the autonomous vehicle prototype Verdino is introduced. Two navigation levels are considered. In the first level, a trajectory is generated from the current position toward a goal that considers two different approaches. In the first, the minimum cost path is obtained using a classical approach (used for regular navigation). The second approach is a little more complex, relying on a set of precomputed primitives representing the motion model of the vehicle, which are used as part of an ARA* algorithm in order to find the best trajectory. This trajectory consists of both forward and backward motion segments for complex maneuvers. In the second level, a local planner is in charge of computing the commands sent to the vehicle in order to follow the trajectory. A set of tentative local trajectories is computed in the Frenet space and scored using several factors, described in this paper. Some results for the two navigation levels are shown at the end of this document. For the global planner, several examples of the maneuvers obtained are shown and certain related factors are quantified and compared. As for the local planner, a study on the influence of the defined weights on the vehicles final behavior is presented. Also, from these tests several configurations have been chosen and ranked according to two different proposed behaviors. The navigation system shown has been tested both in simulated and in real conditions, and the attached video shows the vehicles real-world performance. Graphical abstractDisplay Omitted HighlightsA navigation system has been developed for an autonomous vehicle.The autonomous vehicle can navigate along unstructured and crowded environments.Two planning levels are used, considering two approaches for the first one.The system has been successfully tested in real conditions, results shown.

Safe and Reliable Path Planning for the Autonomous Vehicle Verdino

This paper introduces a local planner which computes a set of commands, allowing an autonomous vehicle to follow a given trajectory. To do so, the platform relies on a localization system, a map and a cost map which represents the obstacles in the environment. The presented method computes a set of tentative trajectories, using a schema based on a Frenet frame obtained from the global planner. These trajectories are then scored using a linear combination of weighted cost functions. In the presented approach, new weights are introduced in order to satisfy the specificities of our autonomous platform, Verdino. A study on the influence of the defined weights in the final behavior of the vehicle is introduced. From these tests, several configurations have been chosen and ranked according to two different proposed behaviors. The method has been tested both in simulation and in real conditions.

FOUR ROTOR HELICOPTER CONTROL LABORATORY PLANT

Abstract In this paper, a test-bed for teaching in multivariable system is presented. Firstly, the different aspects of the prototype construction will be described, making a special emphasis in the mechanics, the design of the sensorial and actuation systems and the prototype control. Next, the real time control software will be explained. The mathematical model of the plant is presented in order to design a control strategy, test it in simulation and validate in the real system.

A Combined Voxel and Particle Filter-Based Approach for Fast Obstacle Detection and Tracking in Automotive Applications

In this paper, a new method for real-time detection, motion estimation, and tracking of generic obstacles using just a 3-D point cloud and odometry information as input is presented. In this approach, a simplification of the world is done using voxels, supported by a particle filter-based 3-D object segmentation and a motion estimation scheme. That combination of techniques leverages a fast and reliable object detection, providing also motion speed and direction information. Four detailed studies have been performed in order to assess the suitability of the method, two of them related to the parameterization of the method and its input point cloud. Another one compares the tracking and detection results with other state-of-the-art methods. Last tests are intended for the characterization of the execution times required. Results are encouraging, with a high detection rate, low error rate, and real-time capable computing performance. In the attached video, it is possible to observe the behavior of the method, both using a stereovision and a light-detection and ranging generated point clouds as an input.

Fast Object Motion Estimation Based on Dynamic Stixels.

The stixel world is a simplification of the world in which obstacles are represented as vertical instances, called stixels, standing on a surface assumed to be planar. In this paper, previous approaches for stixel tracking are extended using a two-level scheme. In the first level, stixels are tracked by matching them between frames using a bipartite graph in which edges represent a matching cost function. Then, stixels are clustered into sets representing objects in the environment. These objects are matched based on the number of stixels paired inside them. Furthermore, a faster, but less accurate approach is proposed in which only the second level is used. Several configurations of our method are compared to an existing state-of-the-art approach to show how our methodology outperforms it in several areas, including an improvement in the quality of the depth reconstruction.

Non-rigid contour flow detection with static cameras for path planning applications

In this paper, a new approach for non rigid obstacle detection and tracking is proposed. Traditionally, this task is performed for each obstacle as a rigid body without considering the local movements of its parts. The presented method combines foreground segmentation techniques for static cameras with nonrigid point set registration algorithms with the objective of having information about the local movements of pedestrians. This information will be used by an electrical unmanned vehicle that will be working inside a closed bioclimatic urbanization in order to perform a more intelligent path planning. This paper has been focused on pedestrian detection, but as no model is used, it can be applied to any type of obstacle. At the end of the paper, results of some tests about the different evaluated algorithms are shown, as well as the final results of all parts of the method working together.