Semi-supervised learning based parking space detection method of automatic encoder in automatic parking

Abstract

The invention provides a semi-supervised learning based parking space detection method of an automatic encoder in automatic parking. The method comprises the following steps that a vehicle-mounted ultrasonic sensor collects parking space detection data; a data preprocessing module constructs an initial semi-supervised learning automatic encoder structure; and in a model training stage, model test is carried out, and a parking space boundary detection coordinate determination module is included. The encoder adopts four target constraints, namely a weight attenuation constraint, a sparse regularization constraint, a reconstruction loss constraint and a label constraint. The model can obtain more unique and local features under unsupervised learning; in addition, under supervision, output and expected output feedback can be generated, feature subordinate categories are effectively extracted, and the optimization progress is adjusted. The encoder can extract effective features in a data set, classifies boundary points with high accuracy and recall rate through a classification network, and finally determines left and right boundaries of the parking space at the highest density of the boundary points. Therefore, the demand of the parking space boundary can be accurately determined.

Description

technical field [0001] The present invention is applied to the field of automatic parking, and relates to technologies such as vehicle-mounted ultrasonic sensor data acquisition and processing, artificial intelligence, and mathematical modeling, especially the use of semi-supervised learning autoencoders to extract the essential boundary features and density average math of ultrasonic sensors. Modeling technology to determine the boundaries of parking spaces is one of the effective methods for parking space detection, and it is the premise of automatic parking follow-up path planning and road strength tracking. Background technique [0002] Today, with the development of the Industry 4.0 era, artificial intelligence technology has been widely used in all walks of life, and automation will gradually become the necessity of social industrial development. In the field of smart cars, also due to the rapid development of computer technology, cars continue to develop into networke...

Images