Henri Bouma

Bio: Henri Bouma is an academic researcher from Netherlands Organisation for Applied Scientific Research. The author has contributed to research in topics: Computer science & Pixel. The author has an hindex of 16, co-authored 82 publications receiving 812 citations. Previous affiliations of Henri Bouma include Philips & Eindhoven University of Technology.

Automatic Detection of Pulmonary Embolism in CTA Images

Abstract: Pulmonary embolism (PE) is a common life-threatening disorder for which an early diagnosis is desirable. We propose a new system for the automatic detection of PE in contrast-enhanced CT images. The system consists of candidate detection, feature computation and classification. Candidate detection focusses on the inclusion of PE-even complete occlusions-and the exclusion of false detections, such as tissue and parenchymal diseases. Feature computation does not only focus on the intensity, shape and size of an embolus, but also on locations and the shape of the pulmonary vascular tree. Several classifiers have been tested and the results show that the performance is optimized by using a bagged tree classifier with two features based on the shape of a blood vessel and the distance to the vessel boundary. The system was trained on 38 CT data sets. Evaluation on 19 other data sets showed that the system generalizes well. The sensitivity of our system on the evaluation data is 63% at 4.9 false positives per data set, which allowed the radiologist to improve the number of detected PE by 22%.

Selection of negative samples and two-stage combination of multiple features for action detection in thousands of videos

Abstract: In this paper, a system is presented that can detect 48 human actions in realistic videos, ranging from simple actions such as `walk' to complex actions such as `exchange'. We propose a method that gives a major contribution in performance. The reason for this major improvement is related to a different approach on three themes: sample selection, two-stage classification, and the combination of multiple features. First, we show that the sampling can be improved by smart selection of the negatives. Second, we show that exploiting all 48 actions' posteriors by two-stage classification greatly improves its detection. Third, we show how low-level motion and high-level object features should be combined. These three yield a performance improvement of a factor 2.37 for human action detection in the visint.org test set of 1,294 realistic videos. In addition, we demonstrate that selective sampling and the two-stage setup improve on standard bag-of-feature methods on the UT-interaction dataset, and our method outperforms state-of-the-art for the IXMAS dataset.

Automatic Detection of Small Surface Targets with Electro-Optical Sensors in a Harbor Environment

Abstract: In modern warfare scenarios naval ships must operate in coastal environments These complex environments, in bays and narrow straits, with cluttered littoral backgrounds and many civilian ships may contain asymmetric threats of fast targets, such as rhibs, cabin boats and jet-skis Optical sensors, in combination with image enhancement and automatic detection, assist an operator to reduce the response time, which is crucial for the protection of the naval and land-based supporting forces In this paper, we present our work on automatic detection of small surface targets which includes multi-scale horizon detection and robust estimation of the background intensity To evaluate the performance of our detection technology, data was recorded with both infrared and visual-light cameras in a coastal zone and in a harbor environment During these trials multiple small targets were used Results of this evaluation are shown in this paper

Real-time tracking and fast retrieval of persons in multiple surveillance cameras of a shopping mall

Abstract: The capability to track individuals in CCTV cameras is important for e.g. surveillance applications at large areas such as train stations, airports and shopping centers. However, it is laborious to track and trace people over multiple cameras. In this paper, we present a system for real-time tracking and fast interactive retrieval of persons in video streams from multiple static surveillance cameras. This system is demonstrated in a shopping mall, where the cameras are positioned without overlapping fields-of-view and have different lighting conditions. The results show that the system allows an operator to find the origin or destination of a person more efficiently. The misses are reduced with 37%, which is a significant improvement.

Re-identification of persons in multi-camera surveillance under varying viewpoints and illumination

Abstract: The capability to track individuals in CCTV cameras is important for surveillance and forensics alike. However, it is laborious to do over multiple cameras. Therefore, an automated system is desirable. In literature several methods have been proposed, but their robustness against varying viewpoints and illumination is limited. Hence performance in realistic settings is also limited. In this paper, we present a novel method for the automatic re-identification of persons in video from surveillance cameras in a realistic setting. The method is computationally efficient, robust to a wide variety of viewpoints and illumination, simple to implement and it requires no training. We compare the performance of our method to several state-of-the-art methods on a publically available dataset that contains the variety of viewpoints and illumination to allow benchmarking. The results indicate that our method shows good performance and enables a human operator to track persons five times faster.

Hierarchical Clustering Multi-Task Learning for Joint Human Action Grouping and Recognition

Abstract: This paper proposes a hierarchical clustering multi-task learning (HC-MTL) method for joint human action grouping and recognition. Specifically, we formulate the objective function into the group-wise least square loss regularized by low rank and sparsity with respect to two latent variables, model parameters and grouping information, for joint optimization. To handle this non-convex optimization, we decompose it into two sub-tasks, multi-task learning and task relatedness discovery. First, we convert this non-convex objective function into the convex formulation by fixing the latent grouping information. This new objective function focuses on multi-task learning by strengthening the shared-action relationship and action-specific feature learning. Second, we leverage the learned model parameters for the task relatedness measure and clustering. In this way, HC-MTL can attain both optimal action models and group discovery by alternating iteratively. The proposed method is validated on three kinds of challenging datasets, including six realistic action datasets (Hollywood2, YouTube, UCF Sports, UCF50, HMDB51 $\&$ UCF101), two constrained datasets (KTH $\&$ TJU), and two multi-view datasets (MV-TJU $\&$ IXMAS). The extensive experimental results show that: 1) HC-MTL can produce competing performances to the state of the arts for action recognition and grouping; 2) HC-MTL can overcome the difficulty in heuristic action grouping simply based on human knowledge; 3) HC-MTL can avoid the possible inconsistency between the subjective action grouping depending on human knowledge and objective action grouping based on the feature subspace distributions of multiple actions. Comparison with the popular clustered multi-task learning further reveals that the discovered latent relatedness by HC-MTL aids inducing the group-wise multi-task learning and boosts the performance. To the best of our knowledge, ours is the first work that breaks the assumption that all actions are either independent for individual learning or correlated for joint modeling and proposes HC-MTL for automated, joint action grouping and modeling.

Statistical optimization for geometric computation : theory and practice

Abstract: Introduction - The aims of this book the features of this book organization and background the analytical mind: strengh and weakness. Fundamentals of linear algebra - Vector and matrix calculus Eigenvalue problem linear systems and optimization matrix and tensor algebra. Probabilities and statistical estimation - probability distributions manifolds and local distributions gaussian distributions and X2 distributions statistical estimation for gaussian models general statistical estimation maximum likelihood estimation Akaike information criterion. Representation of geometric objects - image points and image lines space points and space lines space planes conics space conics and quadrics coordinate transformation and projection. Geometric correction - general theory correction of image points and image lines correction of space points and space lines correction of space planes orthogonality correction conic incidence correction. 3-D computation by stereo vision - epipolar constraint optimal correction of correspondence 3-D reconstruction of points 3-D reconstruction of lines optimal back projection onto a space plane scenes infinitely far away camera calibration errors. Parametric fitting - general theory optimal fitting for image points optimal fitting for image lines optimal fitting for space points optimal fitting for space lines optimal fitting for space planes. Optimal filter - general theory iterative estimation scheme effective gradient approximation reduction from the klaman filter estimation from linear hypotheses. Renormalization - eigenvector fit unbiased eigenvector generalized eigenvalue fit renormalization lincarization second order renormalization. Applications of geometric estimation - image line fitting conic fitting space plane fitting by range sensing space plane fitting by stereo vision. 3-D motion analysis - general theory lincarization and renormalization optimal correction and decomposition reliability of 3-D reconstruction critical surfaces 3-D reconstruction from planar surface motion camera rotation and information. 3-D interpretation of optical flow - optical flow detection theoretical basis of 3-D interpretation optimal estimation of motion parameters. (Part contents).

A Comprehensive Survey of Vision-Based Human Action Recognition Methods.

Abstract: Although widely used in many applications, accurate and efficient human action recognition remains a challenging area of research in the field of computer vision. Most recent surveys have focused on narrow problems such as human action recognition methods using depth data, 3D-skeleton data, still image data, spatiotemporal interest point-based methods, and human walking motion recognition. However, there has been no systematic survey of human action recognition. To this end, we present a thorough review of human action recognition methods and provide a comprehensive overview of recent approaches in human action recognition research, including progress in hand-designed action features in RGB and depth data, current deep learning-based action feature representation methods, advances in human⁻object interaction recognition methods, and the current prominent research topic of action detection methods. Finally, we present several analysis recommendations for researchers. This survey paper provides an essential reference for those interested in further research on human action recognition.

Learning Deep Representations for Ground to Aerial Geolocalization (Open Access)

Abstract: : The recent availability of geo-tagged images and rich geospatial data has inspired a number of algorithms for image based geolocalization. Most approaches predict the location of a query image by matching to ground-level images with known locations (e.g., street-view data). However, most of the Earth does not have ground-level reference photos available. Fortunately, more complete coverage is provided by oblique aerial or bird's eye imagery. In this work, we localize a ground-level query image by matching it to a reference database of aerial imagery. We use publicly available data to build a dataset of 78K aligned crossview image pairs. The primary challenge for this task is that traditional computer vision approaches cannot handle the wide baseline and appearance variation of these cross-view pairs. We use our dataset to learn a feature representation in which matching views are near one another and mismatched views are far apart. Our proposed approach, Where-CNN, is inspired by deep learning success in face verification and achieves significant improvements over traditional hand-crafted features and existing deep features learned from other large-scale databases. We show the effectiveness of Where-CNN in finding matches between street view and aerial view imagery and demonstrate the ability of our learned features to generalize to novel locations.