Fatemeh Karimi Nejadasl - Academia.edu (original) (raw)

Papers by Fatemeh Karimi Nejadasl

The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Apr 26, 2013

This paper addresses the problem of vehicle detection from an image sequence in difficult cases. ... more This paper addresses the problem of vehicle detection from an image sequence in difficult cases. Difficulties are notably caused by relatively small vehicles, vehicles that appear with low contrast or vehicles that drive at low speed. The image sequence considered here is recorded by a hovering helicopter and was stabilized prior to the vehicle detection step considered here. A practical algorithm is designed and implemented for this purpose of vehicle detection. Each pixel is identified firstly as either a background (road) or a foreground (vehicle) pixel by analyzing its gray-level temporal profile in a sequential way. Secondly, a vehicle is identified as a cluster of foreground pixels. The results of this new method are demonstrated on a test image-sequence featuring very congested traffic but also smoothly flowing traffic. It is shown that for both traffic situations the method is able to successfully detect low contrast, small size and low speed vehicles.

arXiv (Cornell University), Mar 29, 2021

Not all video frames are equally informative for recognizing an action. It is computationally inf... more Not all video frames are equally informative for recognizing an action. It is computationally infeasible to train deep networks on all video frames when actions develop over hundreds of frames. A common heuristic is uniformly sampling a small number of video frames and using these to recognize the action. Instead, here we propose full video action recognition and consider all video frames. To make this computational tractable, we first cluster all frame activations along the temporal dimension based on their similarity with respect to the classification task, and then temporally aggregate the frames in the clusters into a smaller number of representations. Our method is end-to-end trainable and computationally efficient as it relies on temporally localized clustering in combination with fast Hamming distances in feature space. We evaluate on UCF101, HMDB51, Breakfast, and Something-Something V1 and V2, where we compare favorably to existing heuristic frame sampling methods.

Journal of Synchrotron Radiation, Nov 29, 2012

The typical dose used to record cryo-electron microscopy images from vitrified biological specime... more The typical dose used to record cryo-electron microscopy images from vitrified biological specimens is so high that radiation-induced structural alterations are bound to occur during data acquisition. Integration of all scattered electrons into one image can lead to significant blurring, particularly if the data are collected from an unsupported thin layer of ice suspended over the holes of a support film. Here, the dose has been fractioned and exposure series have been acquired in order to study beam-induced specimen movements under low dose conditions, prior to bubbling. Gold particles were added to the protein sample as fiducial markers. These were automatically localized and tracked throughout the exposure series and showed correlated motions within small patches, with larger amplitudes of motion vectors at the start of a series compared with the end of each series. A non-rigid scheme was used to register all images within each exposure series, using natural neighbor interpolation with the gold particles as anchor points. The procedure increases the contrast and resolution of the examined macromolecules.

Sensors, Sep 5, 2014

In this paper, we propose an automatic and sequential method for the registration of an image seq... more In this paper, we propose an automatic and sequential method for the registration of an image sequence of a road area without ignoring scene-induced motion. This method contributes to a larger work, aiming at vehicle tracking. A typical image sequence is recorded from a helicopter hovering above the freeway. The demand for automation is inevitable due to the large number of images and continuous changes in the traffic situation and weather conditions. A framework is designed and implemented for this purpose. The registration errors are removed in a sequential way based on two homography assumptions. First, an approximate registration is obtained, which is efficiently refined in a second step, using a restricted search area. The results of the stabilization framework are demonstrated on an image sequence consisting of 1500 images and show that our method allows a registration between arbitrary images in the sequence with a geometric error of zero in pixel accuracy.

arXiv (Cornell University), Jul 24, 2020

Semantic segmentation is an important component in the perception systems of autonomous vehicles.... more Semantic segmentation is an important component in the perception systems of autonomous vehicles. In this work, we adopt recent advances in both image and point cloud segmentation to achieve a better accuracy in the task of segmenting LiDAR scans. KPRNet improves the convolutional neural network architecture of 2D projection methods and utilizes KPConv to replace the commonly used post-processing techniques with a learnable point-wise component which allows us to obtain more accurate 3D labels. With these improvements our model outperforms the current best method on the SemanticKITTI benchmark, reaching an mIoU of 63.1.

The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 2018

One of the biggest challenges for an autonomous vehicle (and hence the WEpod) is to see the world... more One of the biggest challenges for an autonomous vehicle (and hence the WEpod) is to see the world as humans would see it. This understanding is the base for a successful and reliable future of autonomous vehicles. Real-world data and semantic segmentation generally are used to achieve full understanding of its surroundings. However, deploying a pretrained segmentation network to a new, previously unseen domain will not attain similar performance as it would on the domain where it is trained on due to the differences between the domains. Although research is done concerning the mitigation of this domain shift, the factors that cause these differences are not yet fully explored. We filled this gap with the investigation of several factors. A base network was created by a two-step finetuning procedure on a convolutional neural network (SegNet) which is pretrained on CityScapes (a dataset for semantic segmentation). The first tuning step is based on RobotCar (road scenery dataset recorded in Oxford, UK) while afterwards this network is fine-tuned for a second time but now on the KITTI (road scenery dataset recorded in Germany) dataset. With this base, experiments are used to obtain the importance of factors such as horizon line, colour and training order for a successful domain adaptation. In this case the domain adaptation is from the KITTI and RobotCar domain to the WEpod domain. For evaluation, groundtruth labels are created in a weakly-supervised setting. Negative influence was obtained for training on greyscale images instead of RGB images. This resulted in drops of IoU values up to 23.9% for WEpod test images. The training order is a main contributor for domain adaptation with an increase in IoU of 4.7%. This shows that the target domain (WEpod) is more closely related to RobotCar than to KITTI.

To increase robustness in registration of image sequences, we investigate a featureless method. T... more To increase robustness in registration of image sequences, we investigate a featureless method. This paper formulates the registration problem as an optimization of an energy function between a reference image and a transformed of target image. A result parameters are estimated using a global optimizer, Differential Evolution, followed by a local optimizer, Nelder-Mead. Our experiments show that the proposed algorithm perform, robustly in a large variety of image content from the road almost empty surrounding to more cluttered one and from simple road shape to more complex.

The Data Collection System A system is defined to collect many long vehicle trajectories under th... more The Data Collection System A system is defined to collect many long vehicle trajectories under the uniform conditions. The elements and parameters of the data collection system are studied in detail in Chapter 2.

In this paper, we address the registration of two images as an optimization problem within indica... more In this paper, we address the registration of two images as an optimization problem within indicated bounds. Our contribution is to identify such situations where the optimum value represents the real transformation parameters between the two images. Consider for example Mean Square Error (MSE) as the energy function: Ideally, a minimum in MSE corresponds to transformation parameters that represent the real transformation between two images. In this paper we demonstrate in which situations the optimum value represents the real transformation parameters between the two images. To quantify the amount of disturbances allowed, these disturbances are simulated for two separate cases: moving objects and illumination variation. The results of the simulation demonstrate the robustness of stabilizing image sequences by means of MSE optimization. Indeed, it is shown that even a large amount of disturbances will not cause the optimization method to fail to find the real solution. Fortunately, the maximal amount of disturbances allowed is larger than the amount of signal disturbances that is typically met in practice.

Radiation damage in single-particle cryo-electron microscopy: effects of dose and dose rate

ArXiv, 2020

Measuring positions, velocities and accelerations/decelerations of individual vehicles in congest... more Measuring positions, velocities and accelerations/decelerations of individual vehicles in congested traffic with standard traffic monitoring equipment, such as inductive loops, are not feasible. The behavior of drivers in the different traffic situations, as re-quired for microscopic traffic flow models, is still not sufficiently known. Remote sensing and computer vision technology are recently being used to solve this problem. In this study we use video images taken from a helicopter above a fixed point of the highway. We address the problem of tracking the movement of previously detected vehicles through a stabilized video sequence. We combine two approaches, optical flow and matching based tracking, improve them by adding constraints and using scale space. Feature elements, i.e. the corners, lines, regions and outlines of each car, are extracted first. Then, optical-flow is used to find for each pixel in the interior of a car the corresponding pixel in the next image, by insertin...

2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

Institute of Earth Observation and Space System, Delft University of Technology, Kluyverweg 1, 26... more Institute of Earth Observation and Space System, Delft University of Technology, Kluyverweg 1, 2629 HS, Delft, The Netherlands f.KarimiNejadasl, bghgorte@tudelft.nl Transport and planning section, Delft University of Technology, Stevinweg 1, 2628 CN, Delft, The Netherlands SP ...