Deep intelligent driving application-oriented traffic sign recognition algorithm

Abstract

The invention discloses a deep intelligent driving application-oriented traffic sign recognition algorithm, based on consideration of multisource open beta data set translation expansibility and stability by a network, a unified category representation of the multisource open beta data set for traffic signs is built, through a data enhancement strategy randomly clipped based on local context information, expansion of the data set is realized, and a multiscale convolution feature map network and an aggregate network iteration alternate verification training strategy is adopted, a detection network and a recognition network with relatively good performance are obtained, and thus the networks are easy to train and faster in convergence; based on a characteristic of convolution from bottom to top, and through a top-to-down fusion multiscale convolution characteristic network modeling method, the recall ratio of small-sized traffic signs is improved; and design of deeper and more complicated Convnet is not pursued to obtain a higher object recognition rate, but aiming at the characteristics of traffic sign targets, through a contrast experiment, an aggregate network capable of achieving better information flow and better performance is proposed, and efficiency recognition of traffic signs is realized.

Description

technical field [0001] The invention belongs to the technical field of traffic control, and in particular relates to a traffic sign recognition algorithm oriented to deep intelligent driving applications. Background technique [0002] Deep learning is an important branch of artificial intelligence. Artificial intelligence based on deep learning architecture has been widely used in various fields such as computer vision, natural language processing, sensor fusion, biometrics, and autonomous driving. In September 2016, the U.S. Department of Transportation issued a policy on testing and deployment of automated vehicles, establishing the SAE J3016 standard as the global industry reference standard for defining automated or self-driving vehicles, which is used to evaluate six levels (L0-L5 ) of autonomous driving technology. At present, autonomous driving is restricted by factors such as laws and management policies. It will take some time for L4 and L5 autonomous driving vehic...

AI Technical Summary

Problems solved by technology

However, the existing Convnet-based TSR algorithm has some problems: First, traffic signs with the same meaning in different countries have different representations, but most of the above research results are tested and evaluated using public beta data sets, so the scalability and universality are low ;Secondly, they are sensitive to image degradation such as noise and occlusion, and have poor stability; Finally, different activation functions, network parameters, number of network layers, and loss functions of classification layers still consume different computing time, which has a certain impact on timeliness

Images