Research on Visual Perception Technology of Autonomous Driving Based on Improved Convolutional Neural Network

Abstract

Abstract With the continuous development of the world’s scientific and technological level and the continuous progress of society and economy, driverless cars have become the key research targets of the existing automotive industry. Based on the convolutional neural network theory, this paper redesigned the Faster R-CNN network framework to improve the target recognition accuracy and apply it to the unmanned visual perception system. At the same time, the improved algorithm is used to test the cars under different simulated travel conditions. The computer vision system toolbox of MATLAB was used to verify the effectiveness of the algorithm, and the driving scene designer was used to construct multi-sensor fusion and simulated road scenes, and the corresponding visual control system was designed to effectively realize the automatic code generation. The experimental results show that the average detection accuracy of the improved algorithm is higher, the algorithm is more feasible and robust, and it is helpful to promote the development of driverless cars.