Detection of Traffic Jams Using ALISA

Based on collective learning systems theory, ALISA (adaptive learning image and signal analysis) is an adaptive image classification engine that has been designed and tested at the Research Institute for Applied Knowledge Processing (FAW) in Ulm, Germany, at Robert Bosch GmbH in Stuttgart, Germany, and at The George Washington University in Washington, D.C., over the last 5 years. Based on an appropriate set of features, during training, ALISA accumulates an n‐dimensional histogram that estimates the probability density function of the feature space, which becomes the basis for classification during testing. The results of the research reported in this paper suggest that ALISA can be used successfully to detect traffic jams on highways. Based on images captured by a video camera observing different highway traffic conditions, ALISA was trained to recognize and differentiate between steadily flowing traffic and stalled traffic. Several standard features were extracted from preprocessed images based on the differences between successive video frames that were integrated over fairly large receptive fields to reduce differential noise. During testing, ALISA displays a picture of the highway with areas of flowing traffic shown in white and areas of stalled traffic shown as images of the stalled vehicles themselves. Because ALISA classifies every segment of the image independently, even directly adjacent lanes and/or clusters of stalled and flowing traffic are correctly classified. Thus either a summary result (i.e., traffic jam or not) or a more detailed spatial distribution of traffic conditions on the highway can be obtained.