A method for constructing an actual virtual map of the road scene for recognition of dangerous situations in real time

Abstract

Increasing traffic in big cities requires rapid improvement of the transport safety. One possible way to achieve it is by using the computer vision technology and methods of the information processing in real time from a video camera on board the vehicle. However, an image processing with wide-angle and high-resolution cameras to enhance the quality of recognition requires a substantial increase in the volume of the information, and often does not allow identifying dangerous situations in the real time. Our method provides a reconstruction of the video data in two-dimensional coordinate system «actual virtual map» with a smaller amount of information that changes dynamically and can significantly reduce the amount of the computation. Detected moving objects from the road scene are transferred into objects on this map representing the area around the vehicle into consideration. Properties that describe the movement and location of each object with respect to the vehicle in consideration are formed. The presence of such a dynamically formed up-to-date information provides a full picture of the moving objects and their directions of the movement. The method improves the accuracy of the identification of the traffic facilities, reduces the amount of calculation and the percentage of false detection of the objects. Such information also operates as a feedback for the object detection algorithm, determines the vehicle location in each frame and minimizing the search time. The method allows better control of the situation around the vehicle in consideration, as well as the use of cameras with higher resolution and increases the detection accuracy and speed of the objects moving around.