Deep learning-based signal lamp duration detection method

Abstract

The present invention discloses a deep learning-based signal lamp duration detection method and belongs to the field of image processing and pattern recognition. According to the technical scheme of the invention, a vehicle-mounted camera system is adopted to collect the image of a signal lamp. The image of the signal lamp is subjected to size normalization, and the digital of the signal lamp is centered, wherein the size of the image is fixed to be 32 * 32 pixels. A fast feature-implanted convolution structure is adopted to train samples in a training set, so that a training model is obtained. Based on the weight file of the well trained training model, the image of the signal lamp, collected by the vehicle-mounted camera system, is subjected to signal lamp duration detection. Therefore, the detection accuracy of the signal lamp is greatly improved, and the duration of the signal lamp can be detected in real time. In this way, the information of the signal lamp at an intersection can be timely obtained.

Description

technical field [0001] The invention belongs to the field of image processing and pattern recognition, in particular to a method for detecting the duration of signal lamps based on deep learning. [0002] technical background [0003] In recent years, with the acquisition of data, the discovery of advanced theories, and the development of high-performance parallel computing technology, feature learning technology based on deep learning has made breakthroughs in many research fields such as vision, speech, and language. Widely recognized in industry and academia. [0004] Deep learning is a computing model composed of multiple processing layers, which can obtain multi-abstract representations of data through learning. This method improves visual object recognition and detection performance, and many fields benefit from it. Deep learning requires only a small amount of human intervention and is very suitable for current large-scale computing systems and massive data. [0005...

AI Technical Summary

Problems solved by technology

The disadvantage of this method is that the process of searching traffic lights based on distribution information and coordinate network has poor real-time performance and low calculation efficiency, and is not suitable for the detection of traffic lights in real environments.

The disadvantage of this method is that it is greatly affected by the on-board sensors, and the implementation is complicated, so it is not suitable for accurate and fast detection of traffic lights.

Images