摘要 :
With the latest advances in image sensor technology, cameras are able to generate video with tens of megapixels per frame. These high resolution videos streams offer great potential to be used in the surveillance domain. For groun...
展开
With the latest advances in image sensor technology, cameras are able to generate video with tens of megapixels per frame. These high resolution videos streams offer great potential to be used in the surveillance domain. For ground based systems, gigapixel streams are already used with great effect as illustrated by the ICME 2019 crowd counting challenge. However, for Unmanned Aerial Vehicles (UAVs), this vast stream of data exceeds the limit of transmission bandwidth to send this data back to the ground. On board data analysis and selection is thus required to use and benefit from high resolution cameras. This paper presents a result of the CAVIAR project, where a combination of hardware and algorithms was designed to answer the question: "how to exploit a high resolution high frame rate camera on board a UAV?'. With the associated size, weight and power limitations, we implement data reduction by deploying deep learning on hardware to find the relevant information and transmit it to an operator station. The proposed solution aims at employing the high resolution potential of the sensor only onto objects of interest. We encode and transmit the identified regions containing those objects of interest (ROI) at the original resolution and framerate, while also transmitting the downscaled background to provide context for an operator. We demonstrate using a 35 fps, 65 Megapixel camera that this set-up indeed saves considerable bandwidth while retaining all important video data at high quality at the same time.
收起
摘要 :
With the latest advances in image sensor technology, cameras are able to generate video with tens of megapixels per frame. These high resolution videos streams offer great potential to be used in the surveillance domain. For groun...
展开
With the latest advances in image sensor technology, cameras are able to generate video with tens of megapixels per frame. These high resolution videos streams offer great potential to be used in the surveillance domain. For ground based systems, gigapixel streams are already used with great effect as illustrated by the ICME 2019 crowd counting challenge. However, for Unmanned Aerial Vehicles (UAVs), this vast stream of data exceeds the limit of transmission bandwidth to send this data back to the ground. On board data analysis and selection is thus required to use and benefit from high resolution cameras. This paper presents a result of the CAVIAR project, where a combination of hardware and algorithms was designed to answer the question: "how to exploit a high resolution high frame rate camera on board a UAV?'. With the associated size, weight and power limitations, we implement data reduction by deploying deep learning on hardware to find the relevant information and transmit it to an operator station. The proposed solution aims at employing the high resolution potential of the sensor only onto objects of interest. We encode and transmit the identified regions containing those objects of interest (ROI) at the original resolution and framerate, while also transmitting the downscaled background to provide context for an operator. We demonstrate using a 35 fps, 65 Megapixel camera that this set-up indeed saves considerable bandwidth while retaining all important video data at high quality at the same time.
收起
摘要 :
Airborne LiDAR has become the main technology to provide data for Digital Surface Models (DSM) and Digital Elevation Models (DEM) for various purposes, including orthorectification. At the same time, digital line scanners such as ...
展开
Airborne LiDAR has become the main technology to provide data for Digital Surface Models (DSM) and Digital Elevation Models (DEM) for various purposes, including orthorectification. At the same time, digital line scanners such as the ADS provide multiple stereo coverage, which can be used for image-based DSM/DEM derivation. Besides saving the LiDAR acquisition cost, such data provide several advantages over LiDAR results, especially since orthoimage generation can be based on the same data set. In that regard, North West Geomatics and Leica Geosystems have developed a DSM generation tool for ADS data. The underlying approach is Semi-Global Matching (SGM), which is suited for high-performance and high-resolution DSM computation. This paper presents the SGM approach for ADS and compares the results against LiDAR - in terms of data processing as well as DSM/DEM resolution and accuracy. The SGM and LiDAR properties are compared and exemplarily illustrated based on a sub-urban area in Romanshorn, Switzerland. It is shown that SGM can be used as an alternative to LiDAR. For certain applications such as high resolution DSM generation or orthoimage production in general - where it saves the additional flight costs - SGM is even considered the preferred choice.
收起
摘要 :
Airborne LiDAR has become the main technology to provide data for Digital Surface Models (DSM) and Digital Elevation Models (DEM) for various purposes, including orthorectification. At the same time, digital line scanners such as ...
展开
Airborne LiDAR has become the main technology to provide data for Digital Surface Models (DSM) and Digital Elevation Models (DEM) for various purposes, including orthorectification. At the same time, digital line scanners such as the ADS provide multiple stereo coverage, which can be used for image-based DSM/DEM derivation. Besides saving the LiDAR acquisition cost, such data provide several advantages over LiDAR results, especially since orthoimage generation can be based on the same data set. In that regard, North West Geomatics and Leica Geosystems have developed a DSM generation tool for ADS data. The underlying approach is Semi-Global Matching (SGM), which is suited for high-performance and high-resolution DSM computation. This paper presents the SGM approach for ADS and compares the results against LiDAR - in terms of data processing as well as DSM/DEM resolution and accuracy. The SGM and LiDAR properties are compared and exemplarily illustrated based on a sub-urban area in Romanshorn, Switzerland. It is shown that SGM can be used as an alternative to LiDAR. For certain applications such as high resolution DSM generation or orthoimage production in general - where it saves the additional flight costs - SGM is even considered the preferred choice.
收起
摘要 :
This paper describes the algorithm used to further automate the workflow to triangulate images from the multi-line Aerial Digital Sensor ADS from Leica Geosystems. Up to now many parts of the triangulation process have been succes...
展开
This paper describes the algorithm used to further automate the workflow to triangulate images from the multi-line Aerial Digital Sensor ADS from Leica Geosystems. Up to now many parts of the triangulation process have been successfully automated like point measurement, bundle adjustment with automatic blunder removal, automatic variance components estimation and automatic selection of self-calibration parameters. What still requires human interaction is the quality analysis. The key to full automation in triangulation is the quality control. The new process is characterized by a loop, which consists of automatic point measurement, bundle adjustment and automatic quality control. The goal of the process is to obtain the orientation for the block automatically with minimum user interaction. The required quality of the project in terms of accuracy on the ground is specified by the user at project start. The process starts from a sparse tie point pattern, which is successively densified until the requested quality is achieved. Typical for triangulation of line sensors is the use of orientation fixes. In this new approach the time or distance between two orientation fixes is no longer constant. Instead orientation fixes are placed at variable intervals. Those intervals are automatically determined. For quality control each strip is divided into sections, which are defined by the region between two orientation fixes. Each section is further divided into cells. The analysis based on statistical criteria is applied to each cell. To trust in automatically generated results the reliability is important. Therefore the quality criteria are based on external reliability values. As the quality is based on regions on the ground it can be presented in a simple colour coded form. Green means that the requested quality was achieved and red means that this region requires further attention. This way the user is directly guided to those areas where the algorithm could not fulfill the requested criteria. The advantage of the new algorithm is that it works much faster compared to the approach with constant spacing of orientation fixes and very dense point pattern. The old approach was always aiming at best quality, which may not be needed in every project. The new approach will create orientation values with a quality, which is sufficient for the project.
收起
摘要 :
This paper describes the algorithm used to further automate the workflow to triangulate images from the multi-line Aerial Digital Sensor ADS from Leica Geosystems. Up to now many parts of the triangulation process have been succes...
展开
This paper describes the algorithm used to further automate the workflow to triangulate images from the multi-line Aerial Digital Sensor ADS from Leica Geosystems. Up to now many parts of the triangulation process have been successfully automated like point measurement, bundle adjustment with automatic blunder removal, automatic variance components estimation and automatic selection of self-calibration parameters. What still requires human interaction is the quality analysis. The key to full automation in triangulation is the quality control. The new process is characterized by a loop, which consists of automatic point measurement, bundle adjustment and automatic quality control. The goal of the process is to obtain the orientation for the block automatically with minimum user interaction. The required quality of the project in terms of accuracy on the ground is specified by the user at project start. The process starts from a sparse tie point pattern, which is successively densified until the requested quality is achieved. Typical for triangulation of line sensors is the use of orientation fixes. In this new approach the time or distance between two orientation fixes is no longer constant. Instead orientation fixes are placed at variable intervals. Those intervals are automatically determined. For quality control each strip is divided into sections, which are defined by the region between two orientation fixes. Each section is further divided into cells. The analysis based on statistical criteria is applied to each cell. To trust in automatically generated results the reliability is important. Therefore the quality criteria are based on external reliability values. As the quality is based on regions on the ground it can be presented in a simple colour coded form. Green means that the requested quality was achieved and red means that this region requires further attention. This way the user is directly guided to those areas where the algorithm could not fulfill the requested criteria. The advantage of the new algorithm is that it works much faster compared to the approach with constant spacing of orientation fixes and very dense point pattern. The old approach was always aiming at best quality, which may not be needed in every project. The new approach will create orientation values with a quality, which is sufficient for the project.
收起