The method comprises first converting an RGB frame 18 to an HSV frame which is then filtered by at least one threshold value to obtain at least one saturation frame 66. A contour is then extracted from the at least one saturation frame. Accordingly, a first portion of the RGB frame is cropped in order to encompass an area including the at least one contour. The first portion is then classified, for example to determine whether it corresponds to a not-a-traffic-light class, or a red/yellow/green-traffic-light class 10, or the like. The classification could be performed by an artificial neural network or by using affinity scores. The method could be applied to controlling a vehicle or informing the driver of a vehicle (12) from which images can be captured by sensors (16). The efficiency of the classification process is improved by classifying only a cropped region of interest.
Traffic light detection and classification using computer vision and deep learning
2018-07-25
Patent
Elektronische Ressource
Englisch
Deep learning using computer vision in self driving cars for lane and traffic sign detection
BASE | 2019
|Traffic Sign Classification using Deep Learning
IEEE | 2023
|Corrosion detection with computer vision and deep learning
TIBKAT | 2022
|