Frans C. A. Groen

Three-Dimensional Skeletonization: Principle and Algorithm

An algorithm is proposed for skeletonization of 3-D images. The criterion to preserve connectivity is given in two versions: global and local. The latter allows local decisions in the erosion process. A table of the decisions for all possible configurations is given in this paper. The algorithm using this table can be directly implemented both on general purpose computers and on dedicated machinery.

Analysis and control of a rubbertuator arm

Abstract.The control of light-weight compliant robot arms is cumbersome due to the fact that their Coriolis forces are large, and the forces exerted by the relatively weak actuators may change in time as the result of external (e.g. temperature) influences. We describe and analyse the behaviour of a light-weight robot arm, the SoftArm robot. It is found that the hysteretic force-position relationship of the arm can be explained from its structure. This knowledge is used in the construction of a neural-network-based controller. Experiments show that the network is able to control the robot arm accurately after a training session of only a few minutes.

Human chromosome classification based on local band descriptors

In this paper several new techniques for automated chromosome analysis are described: one for piecewise-linear chromosome stretching and projection, two for accurately localizing the centromere and one for two-dimensional local band pattern description. A classification procedure is described that is based upon local band descriptors. Classification results obtained with this method are compared with results obtained with the global band description method (WDD functions). Data sets from two different laboratories are used to investigate the influence of the preparation. Results show the suitability of the local description method in its ability to visualize the image processing technique at the level of the chromosome image.

The Box-Cox metric for nearest neighbour classification improvement

Abstract The Nearest Neighbour rule is a distribution-free classification method. In the literature it has been shown, however, that the performance of the method improves if complex, often locally defined metrics are chosen. In this paper we demonstrate that in many real world discrimination problems comparable or even better classification results can be obtained with a simple global metric based on the Box-Cox transformation. In three case studies this is demonstrated: synthetic data, the IRIS data and real radar data. In the latter experiment, a large reduction in classification error of more than a factor four is achieved. Copyright 1997 Pattern Recognition Society.

Symbol recognition in electrical diagrams using probabilistic graph matching

In this paper a method is proposed to recognize symbols in electrical diagrams based on probabilistic matching. The skeletons of the symbols are represented by graphs. After finding the pose of the graph (orientation, translation, scale) by a bounded search for a minimum error transformation, the observed graph is matched to the class models and the likelihood of the match is calculated. Results are given for computer-generated symbols and hand drawn symbols with and without a template. Error rates range from <1% to 8%.

Requirements for motion estimation in image sequences for traffic applications

In traffic applications research in devices that can increase the safety and comfort of vehicles is an important topic. In this paper a digital rear-view mirror is presented that helps the driver to analyse situations on the road behind hint. From the observations from a camera the optical motion in the image plane can be estimated. Based on this motion estimation the real-world motion of the vehicles behind our car is interpreted. The motion interpretation problem is very sensitive for errors introduced in the motion estimation. This paper describes a method that shows how accurate the motion estimation must be to enable the required accuracy of the motion interpretation. This will be done by means of simulation experiments for characteristic situations (vehicle approaching, retreating or shifting lane).

Vehicle detection with a mobile camera: spotting midrange, distant, and passing cars

In this article, a method is presented for detecting vehicles in image sequences without prior knowledge about the position of the road. A single camera placed in a moving vehicle provides image data. For the detection of midrange and distant vehicles, a combination of three clues is used: shadow, entropy, and horizontal symmetry. To detect passing vehicles, a temporal differencing and projected motion is used. The algorithms are tested by means of many different experiments. These experiments illustrate the robust and accurate performance of both approaches.

Mid-range and distant vehicle detection with a mobile camera

We present a method to detect vehicles in image sequences without pre-knowledge about the position of the road. Image data is provided by a single camera placed in a moving vehicle. To detect mid-range and distant vehicles we use a combination of three different clues. First, we select regions in the image plane that exhibit the characteristics of shadow projected underneath a vehicle. Then, we analyse the entropy and horizontal symmetry of these selected regions. Only those regions that contain enough entropy and symmetry are identified as potential vehicles. We test our algorithm by means of many different experiments, all based on practical video data recorded under different circumstances. These experiments illustrate the robust and accurate performance of our detection method.

Colour based Off-Road Environment and Terrain Type Classification

Terrain classification is an important problem that still remains to be solved for off-road autonomous robot vehicle guidance. Often, obstacle detection systems are used which cannot distinguish between solid obstacles such as rocks or soft obstacles such as tall patches of grass. Terrain classification is needed to prevent that the robot is stopped needlessly by the obstacle detection system. It can also be used to recognize sand roads or other drivable areas. In this paper we present a colour based method to classify typical terrain coverings such as sand, grass or foliage. Using colour recognition outdoors is difficult, because the observed colour of a material is heavily influenced by environment conditions such as the scene composition and illumination. A new approach to colour based classification is presented. It is based on the assumption that images with large similarities in environment related properties such as illumination, materials and geometry also have similar pixel distributions in a colour space. Classification based on a maximum likelihood method with Gaussian mixture models (GMMs) is improved by first distinguishing image sets in the training set that share the same environment state. Because the terrain type colours are modelled separately for each found image set, the influence of changing environment conditions is reduced. Terrain types in a new image are classified with the GMMs of the environment state that is the most similar to it. The results show that our approach is able to classify terrain types in real images with large differences in illumination

3D object recognition from 2D images using geometric hashing

Abstract In this paper, a general technique for model-based recognition is discussed, called Geometric Hashing. Its purpose is to identify an object in the scene, together with its position and orientation. This technique is based on an intensive preprocessing stage, done off-line, where transformation invariant features of the models are indexed into a hash table. This makes the actual recognition particularly efficient. The algorithm stands out for its high inherent parallelism and its ability to deal with occluded scenes. This paper focuses on the use of Geometric Hashing for the case of 3D object recognition from 2D images. An efficient method to represent a 3D model by its 2D projections is proposed. Results are presented of experiments on random data and 3D objects. It has been found that distinguishing between different types of features in a model or scene results in a very efficient implementation of Geometric Hashing using a multidimensional hash table. The filtering ratio of this scheme turns out to be high enough to allow raliable recognition with the corerct feature correspondence between model and scene. The algorithm performed succesfully in dealing with scenes with up to 50% of occlusion and performed at speeds in the order of one second on a SPARC station.