Björn Fröhlich

Semantic segmentation with millions of features: integrating multiple cues in a combined random forest approach

In this paper, we present a new combined approach for feature extraction, classification, and context modeling in an iterative framework based on random decision trees and a huge amount of features. A major focus of this paper is to integrate different kinds of feature types like color, geometric context, and auto context features in a joint, flexible and fast manner. Furthermore, we perform an in-depth analysis of multiple feature extraction methods and different feature types. Extensive experiments are performed on challenging facade recognition datasets, where we show that our approach significantly outperforms previous approaches with a performance gain of more than 15% on the most difficult dataset.

A Fast Approach for Pixelwise Labeling of Facade Images

Facade classification is an important subtask for automatically building large 3d city models. In the following we present an approach for pixel wise labeling of facade images using an efficient Randomized Decision Forest classifier and robust local color features. Experiments are performed with a popular facade dataset and a new demanding dataset of pixel wise labeled images from the Label Me project. Our method achieves high recognition rates and is significantly faster for training and testing than other Methods based on expensive feature transformation techniques.

Will this car change the lane? - Turn signal recognition in the frequency domain

Understanding the intention of other road users is a key requirement for autonomous driving. In this regard, one particularly relevant cue is a flashing turn signal, since it gives an important hint regarding the intended driving direction of another vehicle in the next few seconds. As such, turn signals can be considered as one of the first methods invented for car-to-car communication. In contrast to modern radio-based approaches, turn signals are installed in almost every vehicle. However, only image-based methods are able to detect, recognize and understand those signals. In this paper, we present a new method to recognize turn signals of other vehicles in images. Our approach builds upon a robust vehicle detector and involves three major steps applied to each detected vehicle: light spot detection, feature extraction through FFT-based analysis of the temporal signal behavior at each detected light spot, and AdaBoost classification of the extracted feature set. In our experiments, we use solely virtually-generated data for training and evaluate the proposed approach on a large 30 minute real-world image sequence. Our results indicate competitive performance at real-time speeds.

As Time Goes by—Anytime Semantic Segmentation with Iterative Context Forests

We present a new approach for contextual semantic segmentation and introduce a new tree-based framework, which combines local information and context knowledge in a single model. The method itself is also suitable for anytime classification scenarios, where the challenge is to estimate a label for each pixel in an image while allowing an interruption of the estimation at any time. This offers the application of the introduced method in time-critical tasks, like automotive applications, with limited computational resources unknown in advance. Label estimation is done in an iterative manner and includes spatial context right from the beginning. Our approach is evaluated in extensive experiments showing its state-of-the-art performance on challenging street scene datasets with anytime classification abilities.

Large-scale gaussian process multi-class classification for semantic segmentation and facade recognition

This paper deals with the task of semantic segmentation, which aims to provide a complete description of an image by inferring a pixelwise labeling. While pixelwise classification is a suitable approach to achieve this goal, state-of-the-art kernel methods are generally not applicable since training and testing phase involve large amounts of data. We address this problem by presenting a method for large-scale inference with Gaussian processes. Standard limitations of Gaussian process classifiers in terms of speed and memory are overcome by pre-clustering the data using decision trees. This leads to a breakdown of the entire problem into several independent classification tasks whose complexity is controlled by the maximum number of training examples allowed in the tree leaves. We additionally propose a technique which allows for computing multi-class probabilities by incorporating uncertainties of the classifier estimates. The approach provides pixelwise semantics for a wide range of applications and different image types such as those from scene understanding, defect localization, and remote sensing. Our experiments are performed with a facade recognition application that shows the significant performance gain achieved by our method compared to previous approaches.

Efficient Gaussian process classification using random decision forests

Gaussian Processes are powerful tools in machine learning which offer wide applicability in regression and classification problems due to their non-parametric and non-linear behavior. However, one of their main drawbacks is the training time complexity which scales cubically with the number of samples. Our work addresses this issue by combining Gaussian Processes with Randomized Decision Forests to enable fast learning. An important advantage of our method is its simplicity and the ability to directly control the trade-off between classification performance and computation speed. Experiments on an indoor place recognition task show that our method can handle large training sets in reasonable time while retaining a good classification accuracy.

Large-Scale Gaussian Process Inference with Generalized Histogram Intersection Kernels for Visual Recognition Tasks

We present new methods for fast Gaussian process (GP) inference in large-scale scenarios including exact multi-class classification with label regression, hyperparameter optimization, and uncertainty prediction. In contrast to previous approaches, we use a full Gaussian process model without sparse approximation techniques. Our methods are based on exploiting generalized histogram intersection kernels and their fast kernel multiplications. We empirically validate the suitability of our techniques in a wide range of scenarios with tens of thousands of examples. Whereas plain GP models are intractable due to both memory consumption and computation time in these settings, our results show that exact inference can indeed be done efficiently. In consequence, we enable every important piece of the Gaussian process framework—learning, inference, hyperparameter optimization, variance estimation, and online learning—to be used in realistic scenarios with more than a handful of data.

Large-scale Gaussian process classification using random decision forests

Gaussian processes are powerful modeling tools in machine learning which offer wide applicability for regression and classification tasks due to their non-parametric and non-linear behavior. However, one of their main drawbacks is the training time complexity which scales cubically with the number of examples. Our work addresses this issue by combining Gaussian processes with random decision forests to enable fast learning. An important advantage of our method is its simplicity and the ability to directly control the tradeoff between classification performance and computational speed. Experiments on an indoor place recognition task and on standard machine learning benchmarks show that our method can handle large training sets of up to three million examples in reasonable time while retaining good classification accuracy.

Is it Safe to Change the Lane? — Visual Exploration of Adjacent Lanes for Autonomous Driving

Lane changes on multi-lane roads are an important and complex task for autonomous driving because the system has to be sure that the adjacent lane is not occupied by any other object. Existing radar-based systems can be complemented by vision-based methods to increase their reliability. This work presents new methods based on multiple pattern recognition strategies, such as image categorization, applied to serially-produced, side-mirror mounted fish-eye cameras. The focus is on appearance-based methods, such as tire detection and structure analysis, and motion-based methods, such as optical flow. Extensive experiments evaluate all presented methods on long video sequences on German highways. The proposed approach is shown to be effective for all kinds of vehicles, all relevant situations, and under varying weather conditions.

Segmentation of microorganism in complex environments

In this paper, we tackle the problem of finding microorganisms in bright field microscopy images, which is an important and challenging step in various tasks, like classifying soil textures. Apart from bacteria or fungi, these images can contain impurities such as sand particles, which increase the difficulty of microbe detection. Following a semantic segmentation approach, where a label is inferred for each pixel, we achieve encouraging classification results on a database containing five different types of microbes. We review and evaluate multiple techniques including segment classification, conditional random field models, and level set approaches.