Rapid traffic light detection algorithm applied to pilotless automobile

Abstract

The invention discloses a rapid traffic light detection algorithm applied to a pilotless automobile. The algorithm comprises the steps that 1, red and green candidate areas are selected based on the value of each channel of each collected image, wherein the candidate areas comprise a plurality of communicated domains; 2, according to the traffic light area positions and sensor data in the previous image, and the height range of a traffic light, the red and green area positions of a current image are predicted, and a prediction area is formed; 3, the shape and color of the communicated domains in the current image are recognized, and the communicated domains which have the same shape and color and exist in the prediction areas are the traffic light areas. According to the rapid traffic light detection algorithm, a camera is used for collecting the previous image at the previous time moment and the current image at the current time moment, a sensor is used for collecting the automobile speeds at the two time moments, automobile steering angles at the two time moments, a time change curve and included angles between the horizontal plane and the camera at the two time moments, and by means of the characteristic that the traffic light height meets the national standard, the traffic light color and shape detection purposes of shrinking the detection range and improving the detection precision are achieved.

Description

Technical field [0001] The invention relates to the field of traffic information detection, in particular to a fast detection algorithm for traffic lights applied to unmanned vehicles. Background technique [0002] Most of the existing traffic light detection and recognition technologies use computer graphics and image processing technologies, machine learning and other technologies. The general idea is nothing more than: After color space conversion, use a certain threshold to extract the areas that may be traffic lights, and then use complex algorithms or machine learning methods to remove a certain degree of non-traffic light areas. The shortcomings of these methods are very obvious. There are often situations where traffic is not detected, or the virtual scene rate is too high. For example, LED display signs on the roadside and red objects under sunlight will all appear as traffic lights. Especially at night, it is even more powerless; at the same time, the algorithm complex...

AI Technical Summary

Problems solved by technology

The disadvantages of these methods are very obvious. Often, traffic cannot be detected, or the false scene rate is too high. For example, LED display signs on the side of the road and red objects under sunlight will be detected as traffic lights. Especially at night, it is powerless; at the same time, the algorithm complexity is too high and the processing time is too long, which is also a fatal shortcoming. After all, in automatic driving equipment, very fast detection and recognition capabilities are required.

[0003] The methods mentioned above are applicable to a relatively general environment. The devices using the above algorithms are only "driving assistance devices", which only serve as reminders and warnings, and cannot replace humans, so they are not suitable for automatic driving. in the car system

Images