Monocular Vehicle Perception Algorithm Based on Improved U-NET

This paper presents a monocular vision perception algorithm for vehicles based on improved U-Net. Vehicle perception is the key to autonomous driving and is used to determine the pose information of surrounding vehicles. Among the vision perception schemes, the monocular vision vehicle perception algorithm based on neural network has the lowest cost, but its accuracy is low due to the lack of monocular image depth information. In this regard, this paper uses ApolloCar3D data set to study U-Net based monocular vision around the vehicle perception algorithm, in the form of point detection to achieve vehicle positioning, and its position perception and correction. The algorithm is based on U-Net network, and the vehicle combination correction algorithm is designed to further improve the accuracy of the perception results. Through the test set test, the perception result of the improved algorithm is better than the public performance of other typical algorithms and has better real-time performance.