CN-121982621-A - Parking lot environment detection method, system, readable storage medium and computer

CN121982621ACN 121982621 ACN121982621 ACN 121982621ACN-121982621-A

Abstract

The invention provides a method, a system, a readable storage medium and a computer for detecting the environment of a parking lot, wherein the method comprises the steps of obtaining an environment data stream of the parking lot through data acquisition equipment; the method comprises the steps of introducing linear variable convolution into a target detection model, replacing an attention module of the target detection model by a star convolution module to obtain a multi-task target detection model, inputting an environment data stream into an image processing model to obtain a boundary feature vector and a content feature vector, optimizing the image processing model by a feature fusion result obtained by fusion of the boundary feature vector and the content feature vector to obtain an image processing optimization model, carrying out model fusion on the multi-task target detection model and the image processing optimization model, and adding a parking space state judgment rule in the model fusion process to obtain the environment detection model, and realizing environment detection of a parking lot through the environment detection model. The method can realize accurate identification and real-time early warning of the parking space occupation and abnormal events.

Inventors

LIU ZIYAO
TAN LEI
WAN CHENGCHANG

Assignees

江西百胜智能科技股份有限公司

Dates

Publication Date: 20260505
Application Date: 20260120

Claims (10)

1. A method for detecting an environment in a parking lot, comprising: Data acquisition is carried out on a parking lot through data acquisition equipment in a fixed acquisition period to obtain an environment data stream of the parking lot, wherein the data acquisition equipment comprises image acquisition equipment, thermal imaging equipment and radar equipment; Constructing a target detection model, introducing linear variable convolution into the target detection model, and replacing an attention module of the target detection model by using a star-shaped convolution module to obtain a multi-task target detection model; Constructing an image processing model, inputting the environmental data stream into the image processing model to obtain a corresponding boundary feature vector and a content feature vector, carrying out feature fusion on the boundary feature vector and the content feature vector, and optimizing the image processing model by the obtained feature fusion result to obtain an image processing optimization model; And carrying out model fusion on the multitasking target detection model and the image processing optimization model, adding a parking space state judgment rule in the model fusion process to obtain an environment detection model, and realizing the environment detection of the parking lot through the environment detection model.
2. The method according to claim 1, wherein the step of acquiring data of the parking lot at a fixed acquisition period by a data acquisition device to obtain an environmental data stream of the parking lot comprises: Data acquisition is carried out on the parking lot through data acquisition equipment in a fixed acquisition period to obtain RGB image data, point cloud data and thermal imaging data of the parking lot; Processing the point cloud data through calibration projection to generate a corresponding depth map, and registering the thermal imaging data by using a spatial registration algorithm to obtain temperature characterization data; And performing data alignment on space-time by utilizing the RGB image data, the depth map and the temperature characterization data to form an environment data stream of the parking lot.
3. The method of claim 1, wherein the steps of constructing a target detection model and introducing a linearly variable convolution into the target detection model, and replacing an attention module of the target detection model with a star convolution module to obtain a multi-tasking target detection model comprise: Constructing a target detection model, generating sampling coordinates corresponding to input features of the target detection model according to the target detection model, generating convolution coordinates based on the size of a convolution kernel and a linear variable convolution algorithm, and carrying out convolution processing on the sampling coordinates and the convolution coordinates to obtain a preliminary optimization model; and replacing the attention module of the preliminary optimization model by using a star convolution module, and normalizing by using a batch normalization substitution layer to obtain the multi-task target detection model.
4. The parking lot environment detection method according to claim 1, wherein the steps of constructing an image processing model, inputting the environment data stream into the image processing model to obtain a corresponding boundary feature vector and content feature vector, feature-fusing the boundary feature vector and the content feature vector, and optimizing the image processing model by the obtained feature fusion result to obtain an image processing optimization model include: Constructing an image processing model, and sequentially convolving the environmental data stream by using the image processing model to obtain a first convolution vector; the first vector is respectively transmitted to two branches for processing to respectively obtain a boundary feature vector and a content feature vector, and the boundary feature vector and the content feature vector are subjected to feature fusion to obtain a final feature vector; and optimizing the final feature vector to the image processing model to obtain an image processing optimization model.
5. The parking lot environment detection method according to claim 3, wherein the calculation formula of the preliminary optimization model is: In the formula, Representing the sample coordinates corresponding to the input features, The convolution coordinates are represented and, To sample the two-dimensional offset of the coordinates, The convolution weights are represented as such, Representing the number of sampling points; the calculation formula of the loss function of the preliminary optimization model is as follows: In the formula, Representing the cross-entropy loss, Representing the number of samples to be taken, A real tag representing the category of the object, Representing the prediction probability of the preliminary optimization model; Indicating the loss of the cross-over ratio, To predict the ratio of the intersection area to the union area of the frame and the real frame, Representing the euclidean distance of the centroid of the real frame and the predicted frame, 、 Representing the width and height of the prediction frame respectively, 、 Representing the width and height of the real frame respectively, 、 The width and height of the smallest bounding box consisting of the predicted box and the real box are represented respectively, Represents the balance coefficient of the balance-wheel, Representing the loss function of the preliminary optimization model, Compensation coefficients representing the cross-entropy loss, And a compensation coefficient representing the cross-ratio loss.
6. A parking lot environment detection system, comprising: The system comprises a data acquisition module, a data acquisition module and a data processing module, wherein the data acquisition module is used for carrying out data acquisition on a parking lot through data acquisition equipment at a fixed acquisition period to obtain an environment data stream of the parking lot, and the data acquisition equipment comprises image acquisition equipment, thermal imaging equipment and radar equipment; The first model construction module is used for constructing a target detection model, introducing linear variable convolution into the target detection model, and replacing an attention module of the target detection model by using the star convolution module so as to obtain a multi-task target detection model; The second model construction module is used for constructing an image processing model, inputting the environment data stream into the image processing model to obtain a corresponding boundary feature vector and a content feature vector, carrying out feature fusion on the boundary feature vector and the content feature vector, and optimizing the image processing model by the obtained feature fusion result to obtain an image processing optimization model; The environment detection module is used for carrying out model fusion on the multitasking target detection model and the image processing optimization model, adding a parking space state judgment rule in the model fusion process to obtain an environment detection model, and realizing the environment detection of the parking lot through the environment detection model.
7. The parking lot environment detection system of claim 6, wherein the data acquisition module is specifically configured to: Data acquisition is carried out on the parking lot through data acquisition equipment in a fixed acquisition period to obtain RGB image data, point cloud data and thermal imaging data of the parking lot; Processing the point cloud data through calibration projection to generate a corresponding depth map, and registering the thermal imaging data by using a spatial registration algorithm to obtain temperature characterization data; And performing data alignment on space-time by utilizing the RGB image data, the depth map and the temperature characterization data to form an environment data stream of the parking lot.
8. The parking lot environment detection system of claim 6, wherein the first model building module is specifically configured to: Constructing a target detection model, generating sampling coordinates corresponding to input features of the target detection model according to the target detection model, generating convolution coordinates based on the size of a convolution kernel and a linear variable convolution algorithm, and carrying out convolution processing on the sampling coordinates and the convolution coordinates to obtain a preliminary optimization model; and replacing the attention module of the preliminary optimization model by using a star convolution module, and normalizing by using a batch normalization substitution layer to obtain the multi-task target detection model.
9. A readable storage medium having stored thereon a computer program, wherein the program when executed by a processor implements the parking lot environment detection method as claimed in any one of claims 1 to 5.
10. A computer comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor, when executing the computer program, implements the method of parking lot environment detection as claimed in any one of claims 1 to 5.

Description

Parking lot environment detection method, system, readable storage medium and computer Technical Field The present invention relates to the field of data processing technologies, and in particular, to a method and a system for detecting an environment of a parking lot, a readable storage medium, and a computer. Background With rapid development of technology and improvement of living standard of people, motor vehicles have become one of the indispensable tools in people's life, so the scale and management complexity of parking lots are continuously improved. Traditional parking lot environment detection relies on manual inspection or an automation system based on a single sensor, and has the following outstanding problems: 1. The data source is single, the environment sensing capability is insufficient, and most of the existing systems still rely on a single type of sensor, such as a video monitoring system based on a visible light camera or a berth state detection system based on ultrasonic waves. The system has obviously reduced performance in the scenes of severe illumination change, night low illumination, rain and fog weather shielding, complex shadow interference and the like, and is difficult to realize stable identification and continuous tracking of targets such as vehicles, pedestrians, obstacles and the like in a parking lot. ; 2. the multi-mode data fusion mechanism is lacking, wherein part of advanced schemes attempt to introduce a plurality of sensors such as a laser radar, a thermal imaging camera and the like, but the data acquired by different sensors have differences in time stamps, a space coordinate system, a data format and update frequency; 3. the detection algorithm is generalized and has limited adaptability, namely a target detection model based on a convolutional neural network, which is widely applied at present, is easy to miss detection and misdetection when the distribution of training data is different from the real, complex and changeable parking lot environment, and has low accuracy in boundary division and category judgment of targets such as densely parked vehicles, irregularly parked non-motor vehicles, temporarily piled cargoes and the like. Disclosure of Invention Based on this, an object of the present invention is to provide a parking lot environment detection method, system, readable storage medium and computer, so as to at least solve the above-mentioned drawbacks. The invention provides a parking lot environment detection method, which comprises the following steps: Data acquisition is carried out on a parking lot through data acquisition equipment in a fixed acquisition period to obtain an environment data stream of the parking lot, wherein the data acquisition equipment comprises image acquisition equipment, thermal imaging equipment and radar equipment; Constructing a target detection model, introducing linear variable convolution into the target detection model, and replacing an attention module of the target detection model by using a star-shaped convolution module to obtain a multi-task target detection model; Constructing an image processing model, inputting the environmental data stream into the image processing model to obtain a corresponding boundary feature vector and a content feature vector, carrying out feature fusion on the boundary feature vector and the content feature vector, and optimizing the image processing model by the obtained feature fusion result to obtain an image processing optimization model; And carrying out model fusion on the multitasking target detection model and the image processing optimization model, adding a parking space state judgment rule in the model fusion process to obtain an environment detection model, and realizing the environment detection of the parking lot through the environment detection model. Further, the step of acquiring data of the parking lot through the data acquisition device in a fixed acquisition period to obtain an environmental data stream of the parking lot includes: Data acquisition is carried out on the parking lot through data acquisition equipment in a fixed acquisition period to obtain RGB image data, point cloud data and thermal imaging data of the parking lot; Processing the point cloud data through calibration projection to generate a corresponding depth map, and registering the thermal imaging data by using a spatial registration algorithm to obtain temperature characterization data; And performing data alignment on space-time by utilizing the RGB image data, the depth map and the temperature characterization data to form an environment data stream of the parking lot. Further, the steps of constructing a target detection model, introducing a linear variable convolution into the target detection model, and replacing an attention module of the target detection model with a star convolution module to obtain a multi-task target detection model include: Constructing a target detection model, gener