Object detection in high resolution images is a new challenge that the remote sensing community is facing thanks to introduction of unmanned aerial vehicles and monitoring cameras. One of the interests is to detect and trace persons in the images. Different from general objects, pedestrians can have different poses and are undergoing constant morphological changes while moving, this task needs an intelligent solution. Fine-tuning has woken up great interest among researchers due to its relevance for retraining convolutional networks for many and interesting applications. For object classification, detection, and segmentation fine-tuned models have shown state-of-the-art performance. In the present work, we evaluate the performance of fine-tuned models with a variation of training data by comparing Faster Region-based Convolutional Neural Network (Faster R-CNN) Inception v2, Single Shot MultiBox Detector (SSD) Inception v2, and SSD Mobilenet v2. To achieve the goal, the effect of varying training data on performance metrics such as accuracy, precision, F1-score, and recall are taken into account. After testing the detectors, it was identified that the precision and recall are more sensitive on the variation of the amount of training data. Under five variation of the amount of training data, we observe that the proportion of 60%-80% consistently achieve highly comparable performance, whereas in all variation of training data Faster R-CNN Inception v2 outperforms SSD Inception v2 and SSD Mobilenet v2 in evaluated metrics, but the SSD converges relatively quickly during the training phase. Overall, partitioning 80% of total data for fine-tuning trained models produces efficient detectors even with only 700 data samples.

No abstract

No abstract

No abstract

Photovoltaic and rainwater harvesting assessment on rooftop has been studied extensively. Detailed methodologies are available over large study areas and designed to use data that are usually difficult and expensive to acquire. However, much less attention has been paid to the use of low-cost data for the estimation of photovoltaic parameters and rainwater collection in individual buildings. In this study, a workflow for extraction of geometrical information used in Photovoltaic and rainwater harvesting potential estimation from UAV optical images used to estimate photovoltaic and rainwater harvesting potential is presented. The optical images captured by the DJI Phantom 4 Unmanned Aerial Vehicle (UAV) were used to compute a point cloud, using state of the art Structure from Motion (SfM) algorithms. The modeling of the roof planes was made based on the spatial relationships between points using a Delaunay triangulation. From the generated model, roof geometrical parameters such as area, slope, and orientation were extracted and compared with reference measurements of Light Detection And Ranging (LiDAR) of the same scene. Statistical results from the experiments show that the SfM and LiDAR extracted parameters are very similar. The geometric parameters derived from UAV optical images can be used to support the analysis of the photovoltaic and rainwater harvesting potential in individual buildings. This method has the advantage to achieve results through the combination of low-cost technologies for data acquisition and processing, resulting in an easily reproducible methodology.