Dominik Alexander Klein

Center-surround divergence of feature statistics for salient object detection

In this paper, we introduce a new method to detect salient objects in images. The approach is based on the standard structure of cognitive visual attention models, but realizes the computation of saliency in each feature dimension in an information-theoretic way. The method allows a consistent computation of all feature channels and a well-founded fusion of these channels to a saliency map. Our framework enables the computation of arbitrarily scaled features and local center-surround pairs in an efficient manner. We show that our approach outperforms eight state-of-the-art saliency detectors in terms of precision and recall.

Adaptive real-time video-tracking for arbitrary objects

In this paper, we present a visual object tracker for mobile systems that is able to specialize to individual objects during tracking. The core of our method is a novel observation model and the way it is automatically adapted to a changing object and background appearance over time. The model is integrated into the well known Condensation algorithm (SIR filter) for statistical inference, and it consists of a boosted ensemble of simple threshold classifiers built upon center-surround Haar-like features, which the filter continuously updates based on the images perceived. We present optimizations and reasonable approximations to limit the computational costs. Thus, the final algorithms are capable of processing video input at real-time. To experimentally investigate the gain of adapting the observation model we compare two different approaches with a non-adapting version of our observation model: maintaining a single observation model for all particles, and maintaining individual observation models for each particle. In addition, experiments were conducted to compare system performances between the proposed algorithms and two other state of the art Condensation based tracking approaches.

A multisize superpixel approach for salient object detection based on multivariate normal distribution estimation.

This paper presents a new method for salient object detection based on a sophisticated appearance comparison of multisize superpixels. Those superpixels are modeled by multivariate normal distributions in CIE-Lab color space, which are estimated from the pixels they comprise. This fitting facilitates an efficient application of the Wasserstein distance on the Euclidean norm (W2) to measure perceptual similarity between elements. Saliency is computed in two ways. On the one hand, we compute global saliency by probabilistically grouping visually similar superpixels into clusters and rate their compactness. On the other hand, we use the same distance measure to determine local center-surround contrasts between superpixels. Then, an innovative locally constrained random walk technique that considers local similarity between elements balances the saliency ratings inside probable objects and background. The results of our experiments show the robustness and efficiency of our approach against 11 recently published state-of-the-art saliency detection methods on five widely used benchmark data sets.

SURE: surface entropy for distinctive 3d features

In this paper, we present SURE features --- a novel combination of interest point detector and descriptor for 3D point clouds and depth images. We propose an entropy-based interest operator that selects distinctive points on surfaces. It measures the variation in surface orientation from surface normals in the local vicinity of a point. We complement our approach by the design of a view-pose-invariant descriptor that captures local surface curvature properties, and we propose optional means to incorporate colorful texture information seamlessly. In experiments, we compare our approach to a state-of-the-art feature detector in depth images (NARF) and demonstrate similar repeatability of our detector. Our novel pair of detector and descriptor achieves superior results for matching interest points between images and also requires lower computation time.

Boosting scalable gradient features for adaptive real-time tracking

Recently, several image gradient and edge based features have been introduced. In unison, they all discovered that object shape is a strong cue for recognition and tracking. Generally their basic feature extraction relies on pixel-wise gradient or edge computation using discrete filter masks, while scale invariance is later achieved by higher level operations like accumulating histograms or abstracting edgels to line segments. In this paper we show a novel and fast way to compute region based gradient features which are scale invariant themselves. We developed specialized, quick learnable weak classifiers that are integrated into our adaptively boosted observation model for particle filter based tracking. With an ensemble of region based gradient features this observation model is able to reliably capture the shape of the tracked object. The observation model is adapted to new object and background appearances while tracking. Thus we developed advanced methods to decide when to update the model, or in other words, if the filter is on target or not. We evaluated our approach using the BoBoT1 as well as the PROST2 datasets.

Salient Pattern Detection Using W 2 on Multivariate Normal Distributions

Saliency is an attribute that is not included in an object itself, but arises from complex relations to the scene. Common belief in neuroscience is that objects are eye-catching if they exhibit an anomaly in some basic feature of human perception. This enables detection of object-like structures without prior knowledge. In this paper, we introduce an approach that models these object-to-scene relations based on probability theory. We rely on the conventional structure of cognitive visual attention systems, measuring saliency by local center to surround differences on several basic feature cues and multiple scales, but innovate how to model appearance and to quantify differences. Therefore, we propose an efficient procedure to compute ML-estimates for (multivariate) normal distributions of local feature statistics. Reducing feature statistics to Gaussians facilitates a closed-form solution for the W 2-distance (Wasserstein metric based on the Euclidean norm) between a center and a surround distribution. On a widely used benchmark for salient object detection, our approach, named CoDi-Saliency (for Continuous Distributions), outperformed nine state-of-the-art saliency detectors in terms of precision and recall.

Adaptive Multi-cue 3D Tracking of Arbitrary Objects

We present a general method for RGB-D data that is able to track arbitrary objects in real-time in challenging real-world scenarios. The method is based on the Condensation algorithm. The observation model consists of a target/background classifier that is boosted from a pool of grayscale, color, and depth features. The training set of the observation model is updated with new examples from tracking and the classifier is re-trained to cope with the new appearances of the target. A mechanism maintains a small set of specialized candidate features in the pool, thus decreasing the computational time, while keeping the performance stable. Depth measurements are integrated into the prediction of the 3D state of the particles. We evaluate our approach with a new benchmark for RGB-D tracking algorithms; the results prove our method to be robust under real-world settings, being able to keep track of the targets over 96% of the time.

Moving pedestrian detection based on motion segmentation

The detection of moving pedestrians is of major importance in the area of robot vision, since information about such persons and their tracks should be incorporated into reliable collision avoidance algorithms. In this paper, we propose a new approach, based on motion analysis, to detect moving pedestrians. Our main contribution is to localize moving objects by motion segmentation on an optical flow field as a preprocessing step, so as to significantly reduce the number of detection windows needed to be evaluated by a subsequent people classifier, resulting in a fast method for real-time systems. Therefore, we align detection windows with segmented motion-blobs using a height-prior rule. Finally, we apply a Histogram of Oriented Gradients (HOG) features based Support Vector Machine with Radial Basis Function kernel (RBF-SVM) to estimate a confidence for each detection window, and thereby locate potential pedestrians inside the segmented blobs. Experimental results on “Daimler mono moving pedestrian detection” benchmark show that our approach obtains a log-average miss rate of 43% in the FPPI range [10-2, 100], which is a clear improvement with respect to the naive HOG+linSVM approach and better than several other state-of-the-art detectors. Moreover, our approach also reduces runtime per frame by an order of magnitude.

Exploring Human Vision Driven Features for Pedestrian Detection

Motivated by the center-surround mechanism in the human visual attention system, we propose to use average contrast maps for the challenge of pedestrian detection in street scenes due to the observation that pedestrians indeed exhibit discriminative contrast texture. Our main contributions are the first to design a local statistical multichannel descriptor to incorporate both color and gradient information. Second, we introduce a multidirection and multiscale contrast scheme based on grid cells to integrate expressive local variations. Contributing to the issue of selecting most discriminative features for assessing and classification, we perform extensive comparisons with respect to statistical descriptors, contrast measurements, and scale structures. By this way, we obtain reasonable results under various configurations. Empirical findings from applying our optimized detector on the INRIA and Caltech pedestrian datasets show that our features yield state-of-the-art performance in pedestrian detection.

Fast moving pedestrian detection based on motion segmentation and new motion features

The detection of moving pedestrians is of major importance for intelligent vehicles, since information about such persons and their tracks should be incorporated into reliable collision avoidance algorithms. In this paper, we propose a new approach to detect moving pedestrians aided by motion analysis. Our main contribution is to use motion information in two ways: on the one hand we localize blobs of moving objects for regions of interest (ROIs) selection by segmentation of an optical flow field in a pre-processing step, so as to significantly reduce the number of detection windows needed to be evaluated by a subsequent people classifier, resulting in a fast method suitable for real-time systems. On the other hand we designed a novel kind of features called Motion Self Difference (MSD) features as a complement to single image appearance features, e. g. Histograms of Oriented Gradients (HOG), to improve distinctness and thus classifier performance. Furthermore, we integrate our novel features in a two-layer classification scheme combining a HOG+Support Vector Machines (SVM) and a MSD+SVM detector. Experimental results on the Daimler mono moving pedestrian detection benchmark show that our approach obtains a log-average miss rate of 36 % in the FPPI range [10−2,100], which is a clear improvement with respect to the naive HOG+SVM approach and better than several other state-of-the-art detectors. Moreover, our approach also reduces runtime per frame by an order of magnitude.