Road Image Enhancement and Real-Time Vehicle Detection at Night Based on Convolutional Neural Network

Authors

  • Tao Dong School of Information Engineering, Liaodong University, Dandong 118000, China

Abstract

In the era of artificial intelligence, machine vision technology has been widely used in security, traffic and other contexts, but the image taken at night by machine vision technology has a low detection effect. Therefore, this paper proposes a nighttime road image enhancement and real-time vehicle detection model based on a convolutional neural network (CNN). For the model, the Retinex theory is applied to build an end-to-end image enhancement structure, and inputs the enhanced image as data into a D-CNN recognition model, which optimizes the Softmax classifier. This approach greatly improves the enhanced image recognition effect. The experimental results show that with the proposed image enhancement model, the MSE value of the proposed R-ETELIIES algorithm is about 500, the performance is the best among the comparison algorithms, and the processing speed is less than 0.1s when processing high-resolution images. Among the recognition and detection models, the recognition rate of the D-CNN model reaches 96%, while the false detection rate is only 5%. The model constructed in this study still has good detection performance in a low-light environment, which has an important influence on the construction of a smart city.

Keywords: convolutional neural network; low light image; image enhancement; linear discriminant analysis; inspection of vehicle

Cite As

T. Dong, "Road Image Enhancement and Real-Time Vehicle Detection at Night Based on Convolutional
Neural Network", Engineering Intelligent Systems, vol. 32 no. 2, pp. 175-184, 2024.






Published

2024-03-01