An expert fuzzy system for predicting object collisions. Its application for avoiding pedestrian accidents

摘要

Collision among moving objects in space is one of the most common risks in daily life. In this context, we have developed an abstract model that allows to detect the presence of risk of future collisions among objects from the video content analysis. Our proposal carries out several stages. First, a camera calibration process calculates the real location of object in scene. Then, we estimate the object speed and their future trajectory in order to predict possible collisions. All the information of the objects is described in an ontology. Using the properties of objects (such as location, speed, trajectories), we have defined a fuzzy rule that permits to identify whether an object is in danger because another could hit him. The use of fuzzy logic results in two points: the collision detection is gradual and the model can be adjusted through membership functions to fuzzy concepts. Furthermore, the proposed model is easily adaptable to any situation and can be applied on various fields. With the aim of testing our proposal, we have focused on pedestrian accidents, a case of special interest since a lot of pedestrians die or are injured in traffic accidents daily. We have developed an application based on our model that is able to predict, in real time, the traffic accidents where a vehicle could run over a pedestrian. The obtained results in the experimental stage show a high performance of the system.