The Experts below are selected from a list of 5463 Experts worldwide ranked by ideXlab platform
Fernández Provecho Nelson - One of the best experts on this subject based on the ideXlab platform.
-
Estimación de velocidad de un UGV usando una cámara monocular
2020Co-Authors: Fernández Provecho NelsonAbstract:[ES] El trabajo presentado en este TFM se centra en el uso de una cámara monocular orientada hacia adelante para estimar la velocidad de un vehículo terrestre y comprobar la precisión de las medidas. El método propuesto procesa el vídeo grabado usando una técnica de "visual odometry" basada en correspondencias 2D-to-2D para obtener una estimación de la posición de la cámara en un fotograma en relación con el anterior usando información de puntos destacables de las dos imágenes. Sin embargo, el vector de translación obtenido es de longitud unitaria, por lo que se ha desarrollado un método para obtener su escala absoluta. Para ello se triangulan puntos de la carretera con dos imágenes consecutivas que permiten obtener un plano de la carretera virtual y medir la distancia de éste a la cámara. Sabiendo la altura real a la que está situada se obtiene un factor de escalado y, junto con el tiempo entre fotogramas, se calcula la velocidad. Los resultados directos del algoritmo tienen bastante ruido y algunos picos de velocidad que no son reales producidos por la presencia de otros vehículos. Esto se resuelve fijando un límite de aceleración máxima y aplicando un filtro de media móvil. La influencia de ese umbral ha sido estudiada y demostrado que es importante a la hora de analizar los resultados finales. Utilizando esta metodología se han conseguido buenas estimaciones, demostrando que es efectiva, dando resultados precisos, especialmente cuando las velocidades no son muy grandes y hay suficientes puntos en la carretera. Además, el método propuesto se ha analizado usando diferentes resoluciones, tanto temporales como espaciales, mostrando que el efecto que tienen sobre los resultados no es muy grande, aunque, en términos generales, los resultados son mejores con las resoluciones más altas.[EN] The work presented in this thesis focuses on the use of a forward-looking monocular camera to estimate the velocity of a ground vehicle and check the accuracy of its measurements. The proposed method processes the recorded video with a 2D-to-2D-based visual odometry approach to obtain an estimation of the camera pose of one frame with respect to the previous one using the information of salient corners on both images. However, the translation vector of this estimation is of unit length so a method to obtain the absolute scale has been also developed. Points on the Road from two consecutive images are triangulated to obtain a virtual Road Plane to then measure the distance from it to the virtual camera. With the knowledge of the actual height of the sensor, a scale factor is computed and, together with the time between frames, the velocity is obtained. These raw measurements are noisy and also have unreal velocity peaks produced by the presence of other vehicles on the Road, among other factors. To resolve this, a limit on the maximum allowed acceleration is set and a smoothing step is done. The influence of this threshold has been studied and demonstrated to be important when analysing the final results. Good estimations are achieved using the proposed methodology, showing that it is effective, giving accurate results, especially when the velocities are not too big and enough features are on the Road. Additionally, it has been tested for different spatial and temporal resolutions, showing that their effect on the results is not very big, although, in general terms, better results have been obtained for the highest resolutions.Fernández Provecho, N. (2017). UGV Velocity Estimation from a Monocular Camera. http://hdl.handle.net/10251/142843TFG
-
Estimación de velocidad de un UGV usando una cámara monocular
'Universitat Politecnica de Valencia', 2020Co-Authors: Fernández Provecho NelsonAbstract:[ES] El trabajo presentado en este TFM se centra en el uso de una cámara monocular orientada hacia adelante para estimar la velocidad de un vehículo terrestre y comprobar la precisión de las medidas. El método propuesto procesa el vídeo grabado usando una técnica de "visual odometry" basada en correspondencias 2D-to-2D para obtener una estimación de la posición de la cámara en un fotograma en relación con el anterior usando información de puntos destacables de las dos imágenes. Sin embargo, el vector de translación obtenido es de longitud unitaria, por lo que se ha desarrollado un método para obtener su escala absoluta. Para ello se triangulan puntos de la carretera con dos imágenes consecutivas que permiten obtener un plano de la carretera virtual y medir la distancia de éste a la cámara. Sabiendo la altura real a la que está situada se obtiene un factor de escalado y, junto con el tiempo entre fotogramas, se calcula la velocidad. Los resultados directos del algoritmo tienen bastante ruido y algunos picos de velocidad que no son reales producidos por la presencia de otros vehículos. Esto se resuelve fijando un límite de aceleración máxima y aplicando un filtro de media móvil. La influencia de ese umbral ha sido estudiada y demostrado que es importante a la hora de analizar los resultados finales. Utilizando esta metodología se han conseguido buenas estimaciones, demostrando que es efectiva, dando resultados precisos, especialmente cuando las velocidades no son muy grandes y hay suficientes puntos en la carretera. Además, el método propuesto se ha analizado usando diferentes resoluciones, tanto temporales como espaciales, mostrando que el efecto que tienen sobre los resultados no es muy grande, aunque, en términos generales, los resultados son mejores con las resoluciones más altas.[EN] The work presented in this thesis focuses on the use of a forward-looking monocular camera to estimate the velocity of a ground vehicle and check the accuracy of its measurements. The proposed method processes the recorded video with a 2D-to-2D-based visual odometry approach to obtain an estimation of the camera pose of one frame with respect to the previous one using the information of salient corners on both images. However, the translation vector of this estimation is of unit length so a method to obtain the absolute scale has been also developed. Points on the Road from two consecutive images are triangulated to obtain a virtual Road Plane to then measure the distance from it to the virtual camera. With the knowledge of the actual height of the sensor, a scale factor is computed and, together with the time between frames, the velocity is obtained. These raw measurements are noisy and also have unreal velocity peaks produced by the presence of other vehicles on the Road, among other factors. To resolve this, a limit on the maximum allowed acceleration is set and a smoothing step is done. The influence of this threshold has been studied and demonstrated to be important when analysing the final results. Good estimations are achieved using the proposed methodology, showing that it is effective, giving accurate results, especially when the velocities are not too big and enough features are on the Road. Additionally, it has been tested for different spatial and temporal resolutions, showing that their effect on the results is not very big, although, in general terms, better results have been obtained for the highest resolutions.Fernández Provecho, N. (2017). UGV Velocity Estimation from a Monocular Camera. Universitat Politècnica de València. http://hdl.handle.net/10251/142843TFG
Sergiu Nedevschi - One of the best experts on this subject based on the ideXlab platform.
-
Obstacle detection based on the hybrid Road Plane under the weak calibration conditions
2008 IEEE Intelligent Vehicles Symposium, 2008Co-Authors: Pangyu Jeong, Sergiu NedevschiAbstract:This paper presents a new obstacle detection method that is achieved on the depth image. This obstacle detection method can be used both in clumsy office environments and in natural environments. In this paper we focus our attention on the obstacle detection in the case of pitch angle variation (weak calibration conditions) due to uneven-Road surface. The proposed obstacle detection has, mainly, two characteristics. One is the time efficiency aspect in the labelling of detected obstacles, which is achieved on the binarized depth image. The contour pixels of the obstacles of the binarized depth image have 3D information. The other is flexibility aspect in Road Plane that is generated by the LDPD (local difference probability distance) Road classification. The shape of the Road Plane reflects the natural Road surface by implementing hybrid Road Plane that consists of rough Road Plane and dedicated Road Plane. The real-time implementation is possible due to the time efficient grouping and labelling. Even obstacle detection is achieved on the depth image. The error of obstacle detection due to the change in height and position of the Road Plane by pitch angle variation is corrected by the proposed hybrid Road Plane.
Jan Salmen - One of the best experts on this subject based on the ideXlab platform.
-
video based roll angle estimation for two wheeled vehicles
IEEE Intelligent Vehicles Symposium, 2011Co-Authors: Marc Schlipsing, Jakob Schepanek, Jan SalmenAbstract:Video-based driver assistance systems are a key component for intelligent vehicles today. Applications for lane detection, traffic sign recognition, and collision avoidance have been successfully deployed in cars and trucks. State-of-the art algorithms rely on machine learning and therefore depend on invariance conditions, e.g. a fixed image perspective. In order to apply current modules in two-wheeled vehicles one needs to determine the roll angle, i.e. the angle between the Road Plane and the slanted vehicle. It can either be used for parametrisation of the algorithms or for rotation of the video image back to a horizontal alignment. Using an inertial measurement unit to acquire this data is unreasonably expensive. We propose a video-based module that estimates the current roll angle based on gradient orientation histograms to overcome this flaw. Due to the visual structure of a traffic scene we are able to derive possible roll angles from the gradient statistics by correlation with learnt data. Analogously, we estimate the roll rate by correlating subsequent image statistics and stabilise both measures within a linear Kalman filter. Experiments on real image data from various test scenarios show high accuracy of the proposed approach. Thus, estimating the roll angle / rate from video only, enables us to employ established video-based assistance modules for two-wheeled vehicles without any additional hardware expense.
Vinod K Pandey - One of the best experts on this subject based on the ideXlab platform.
-
real time obstacle detection by Road Plane segmentation
International Colloquium on Signal Processing and Its Applications, 2013Co-Authors: S Santhanam, V Balisavira, Vinod K PandeyAbstract:Use of computer vision in the automobile industry has been showing extensive growth for past several years. Such camera based systems assist the driver in making decisions with regard to Road safety and host vehicle safety. This paper presents a semantic based Road Plane segmentation technique to detect moving and non-moving obstacles and warn the driver about perils around the automobile. The technique uses color information of the pixels to create primary Road models. Having these models as a basis, dynamic models, for including various Road related anomalies, are also computed. Next, both the models are merged and finally used to classify incoming image pixels as the Road or the obstacles. Application of the technique on several videos, obtained from a moving camera, mounted on a car, resulted in a significantly high detection of moving and non-moving objects in the Region of Interest (RoI). Computational complexity of the technique is very less and it has been ported on a digital signal processor. The technique can be employed in back, side and front cameras to detect the obstacles for the Road and host vehicle safety.
-
real time object detection by Road Plane segmentation technique for adas
Signal-Image Technology and Internet-Based Systems, 2012Co-Authors: V Balisavira, Vinod K PandeyAbstract:Advanced Driver Assistance Systems (ADAS) are used for assisting the drivers by providing advice and warnings when necessary. CTA (Cross Traffic Alert) systems are a subset of ADAS used for detecting objects (viz., cars, trucks, pedestrians, static objects etc) by using one or more moving cameras, mounted on a vehicle. Usually, CTA systems can detect moving objects within region of interest (ROI). These systems have limitations in detecting static objects present in the RoI and often they fail to detect the objects in shadow regions. Moreover, such systems sometimes detect shadows as the objects. This paper presents a histogram back-projection based Road-Plane segmentation technique. Histogram back-projections' are applied on saturation and value channels of the video, to detect moving and non moving objects in the ROI. Robustness to the shadow is achieved by applying a logical operation on the back-projections of the saturation and the value channels of the video. Effectiveness of the technique is evaluated by applying the technique on several videos, captured under different scenarios, and by measuring true negatives and false positives for the objects. The technique is suitable for real time applications and can be employed in automatic back-up assistance during the host vehicle parking, blind spot detection, pedestrian detection, and other camera applications for the detection of the objects.
Pangyu Jeong - One of the best experts on this subject based on the ideXlab platform.
-
Obstacle detection based on the hybrid Road Plane under the weak calibration conditions
2008 IEEE Intelligent Vehicles Symposium, 2008Co-Authors: Pangyu Jeong, Sergiu NedevschiAbstract:This paper presents a new obstacle detection method that is achieved on the depth image. This obstacle detection method can be used both in clumsy office environments and in natural environments. In this paper we focus our attention on the obstacle detection in the case of pitch angle variation (weak calibration conditions) due to uneven-Road surface. The proposed obstacle detection has, mainly, two characteristics. One is the time efficiency aspect in the labelling of detected obstacles, which is achieved on the binarized depth image. The contour pixels of the obstacles of the binarized depth image have 3D information. The other is flexibility aspect in Road Plane that is generated by the LDPD (local difference probability distance) Road classification. The shape of the Road Plane reflects the natural Road surface by implementing hybrid Road Plane that consists of rough Road Plane and dedicated Road Plane. The real-time implementation is possible due to the time efficient grouping and labelling. Even obstacle detection is achieved on the depth image. The error of obstacle detection due to the change in height and position of the Road Plane by pitch angle variation is corrected by the proposed hybrid Road Plane.
