Werner Ritter

An adaptive system for traffic sign recognition

Traffic sign recognition is a primary goal of almost all road environment understanding systems. A vision system for traffic sign recognition was developed by Daimler-Benz Research Center Ulm. The two main modules of the system are detection and verification (recognition). Here regions of possible traffic signs in a color image sequence are first detected before each of them is verified and recognized. In this paper the authors pay attention to the verification and recognition process. The authors present an adaptive approach and emphasize the importance of the adaptability to various road and traffic sign environments. The authors utilize a distance-weighted k-nearest-neighbor classifier for traffic sign recognition and show its equivalence to the kind of radial basis function networks which can be easily integrated into chips. The authors also present a way to evaluate the uncertainty of recognized traffic signs and demonstrate their approach using real images.

Sensorfusion Using Spatio-Temporal Aligned Video and Lidar for Improved Vehicle Detection

This contribution introduces a novel approach to cross-calibrate automotive vision and ranging sensors. The resulting sensor alignment allows the incorporation of multiple sensor data into a detection and tracking framework. Exemplarily, we show how a realtime vehicle detection system, intended for emergency breaking or ACC applications, benefits from the low level fusion of multibeam lidar and vision sensor measurements in discrimination performance and computational complexity

Obstacle detection based on color blob flow

In this paper we present a new robust approach to extract moving objects in traffic scenes. It allows moving obstacles (cars, motorcycles, ...) to be detected from a moving car without any a priori information about the shape or location of these moving objects. This technique is based on the concept of tracking color blobs in image sequences.

Reinforcing the reliability of pedestrian detection in far-infrared sensing

In this contribution we describe a method to improve the reliability of monocular far-infrared pedestrian detection. In contrast to many other solutions using heuristic approaches, we use a statistical approach, i.e., a pixel classification for head detection in comparison with a classifier for body detection to investigate the performance of pedestrian detection. Using a head detector leads to a significant improvement in the detection precision and promises further advantages in combination with a body detector.

Traffic sign recognition in color image sequences

At the Daimler-Benz Research Center in Ulm a system for traffic sign recognition is under development. The task of the system is to detect and interpret traffic signs in colour image sequences; these images are acquired by a camera mounted in a car. The color segmentation of the incoming images is performed with a high order neural network. Based on this color segmented images and using a priori knowledge, hypotheses on image regions containing traffic signs are generated. The kind of traffic sign is also hypothesized. The preselected image regions are further analysed in order to verify or reject the hypothesis. The analysis finally interprets the contents of the traffic signs. The control of the analysis is supported by knowledge on traffic signs and outdoor scenes in general. The whole knowledge is stored in a framebased network.<<ETX>>

A multiple detector approach to low-resolution FIR pedestrian recognition

In this paper we present a recognition scheme, which is both reliable and fast. The scheme comprises the simultaneous harmonized use of three powerful detection algorithms, the hyper permutation network (HPN), a hierarchical contour matching (HCM) algorithm and a cascaded classifier approach. Each algorithm is evaluated separately and afterwards, based on the evaluation results, the fusion of the detection results is performed by a particle filter approach.

Multiple Cue Data Fusion with Particle Filters for Road Course Detection in Vision Systems

In the following we present a robust, real time applicable fusion system for detection of the road course up to 100m. For that, using a particle filter we developed a robust and flexible system which fuses the information from a digital map system and different cues from an image sensor. In addition we present a new image based feature for road surface detection. The system works in realtime

Multiple-cue data fusion with particle filters for vehicle detection in night view automotive applications

In this contribution we present a sensor data fusing concept utilizing particle filters. The investigation aims at the development of a robust and easy to extend approach, capable of combining the information of different sensors. We use the particle filters characteristics and introduce weighting functions that are multiplied during the measurement update stage of the particle filter implementation. The concept is demonstrated in a vehicle detection system that conjoins symmetry detection, tail lamp detection and radar measurements in night vision applications.

Detection and classification of obstacles in night vision traffic scenes based on infrared imagery

Due to the high number of traffic victims at night the detection and classification of obstacles in night scenes is an important goal of vision based driver assistance systems. Especially non-luminescent objects, like humans and animals, are of interest. In our night-vision project, we use a passive far-infrared sensor to achieve this goal. A pixel based statistical classifier, a region-based segmentation algorithm and a polynomial classifier are used to detect and classify objects. The first classifier finds interesting regions with potential objects, the region-based segmentation algorithm is used to resegment those ROIs, and a quadratic polynomial classifier determines the type of the object. The resegmentation module provides an improvement in detection exactness and classification errors.

Detection and Tracking of Multiple Pedestrians in Automotive Applications

We present a method for tracking an unknown and changing number of far away pedestrians in a video stream. Multiple particle filter instances are utilized which track single pedestrians independently from each other. The tracking is guided by a cascade classifier which is integrated into the particle filter framework. In order to be able to detect hardly visible pedestrians and to filter out isolated false positives of the classifier, we developed a detection criterion for particle filters which follows the track-before-detect paradigm. The system nearly works in real time.