Network, control, communication and computing technologies for intelligent transportation systems overview of the special issue

S. M. Amin, A. García-Ortiz, J. R. Wootton

Research output: Contribution to journalArticlepeer-review

3 Scopus citations
Original languageEnglish (US)
Pages (from-to)1-10
Number of pages10
JournalMathematical and Computer Modelling
Issue number4-7
StatePublished - 1995
Externally publishedYes

Bibliographical note

Funding Information:
state-of-the-art application of machine vision to these various aspects which are critical subsets of the overall aims of the ITS and Prometheus projects. Two papers focus on object recognition using rather novel image processing techniques to detect and recognize objects in natural scenes as viewed by cameras mounted inside the car. In the first paper, “Traffic Sign Recognition Using Color Information” by W. Ritter et al. of Ulm Research Centre, Damiler-Benz AG is again part of the European Prometheus research program. The paper places emphasis on reducing processing time. In order to achieve this, attention is given to reducing the search space by using color as a segmentation process. Measured data is mapped into a color descriptor using a polynomial classifier. The individual pixels are clustered into regions using a fast region grower designed as CCC (Color Connected Components). Human recognition is dependent upon shape as well as color; yet, as pointed out by the authors, the concept of shape is difficult to convey to a computer and notoriously difficult to extract from images. In the past, shape has often been derived by extracting edge information and consequently extracting and manipulating Fourier coefficients. Edge information is inherently unreliable and readily corrupted by spatial quantization and video noise in the imagery. Here, Ritter et al. have avoided that pitfall by using a coarse color shape filter which uses normalized size as a descriptor. By combining the relationships of color and shape in a classifier, the object recognition is obtained in terms of traffic signs. The authors acknowledge that their work needs to be extended to multiple image frames and also realize that they might have difficulty with achromatic traffic signs; nevertheless, the objective of achieving fast, real time, robust recognition of traffic signs clearly has the potential of future realization using the novel techniques outlined in this paper. The second paper, “Object Recognition with Constrained Elastic Models” by M. Schwarzinger et al., Institut fur Neuroinformatik, Ruhr-Universitat Bochum, uses a more classical overall structure of a model based recognizer to identify objects in images of natural scenes. The work jointly funded by the German Federal Ministry of Research and Technology (BMFT) and the German automobile industry as part of the Prometheus project undertakes to detect, track, and recognize leading vehicles based upon a rear view. The objective is to ultimately allow the relief of the driver by the automatic driving of the car with the assistance of the automatic recognizer. The researchers have carefully reduced the number of stored models to a very small finite set representative of the features of various generic classes, e.g., passenger cars, vans, trucks, etc. The types and numbers of features per class are also kept small. The features extracted are lines and corners using a Hough transform. The crux of their problem then is reduced to the task of matching the model features to features extracted from the image. This is achieved using an elastic matching algorithm inspired by the elastic net method first applied to the “traveling salesman problem” by Durbin and Willshaw. However, the model is constrained such that it does not deform to objects outside the deformation limits ascribed to a particular class. This novel approach yields remarkable results; the process is computationally expensive. It is interesting that the researchers have not constrained themselves by hardware implementation or real time considerations. They have focused their attention on developing an appropriate set of algorithms to perform the task. Considering the rapid evolution of real-time hardware processing power, the hardware implementation limitations will eventually be resolved. H.-H. Nagel et al. of the Fraunhofer-Institut fur Informations-und Datenverarbeitung in their paper “FhG-Co-Driver: From Map-Guided Automatic Driving by Machine Vision to a Cooperative Driver Support” present a thorough discourse of IITB history of using largely machine vision to maneuver (automatic lateral and longitudinal control of a road vehicle) along a road. The history includes references to state-of-the-art road following. It includes the commissioning on a van of a rapid prototype one camera system and their early success in February 1981 in following lane borders on private roads at velocities of 35 km/hr. It describes their “tedious, unspectacular work” to produce a robust version for experiments. The paper shows the migration to incorporate collision/obstacle avoidance, integration of a commercial road navigation system, and recognition and maneuvering intersections. The incorporation of the latter required exten-

Cite this