CN-121980422-A - Road state detection method, device, equipment and product based on unified model

CN121980422ACN 121980422 ACN121980422 ACN 121980422ACN-121980422-A

Abstract

The application is suitable for the technical field of computers, and provides a road state detection method, device, equipment and product based on a unified model. The unified model provided by the embodiment of the application is an integrated model framework capable of outputting one or more traffic state type judging results, can uniformly process multiple road state judging requirements, and can cover multiple source characteristics of multiple types of target roads, so that more complete characteristics can be covered under a complex scene, more judging bases are provided, the recognition accuracy, the environmental adaptability and the robustness are improved, the technical route and the model stack which need to be independently modeled for different judging tasks in the prior art are avoided, the repeated investment of computing resources is reduced, the system architecture is prevented from being bloated, and the complexity and the cost of operation and maintenance are reduced.

Inventors

WU YIMENG
LIU GUOPING
CUI PENGFEI
WANG CHANG
WANG HONGYI

Assignees

滴图(北京)科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260130

Claims (10)

1. The road state detection method based on the unified model is characterized by comprising the following steps of: carrying out data extraction on traffic data to obtain multi-source characteristics of a target road; And determining a judgment result of the target road about a traffic state type according to the multi-source characteristics of the target road, wherein the traffic state type comprises at least any one of a road traffic state of the target road, a road section traffic state of a road section to which the target road belongs, a congestion state of the target road and an access control state of the target road.
2. The method of claim 1, wherein the data extracting the traffic data to obtain the multi-source characteristic of the target link comprises: extracting traffic data to obtain various characteristics of a target road; And splicing and combining various characteristics of the target road to obtain the multi-source characteristics of the target road.
3. The method of claim 2, wherein the traffic data comprises at least any one of basic dynamic traffic data, basic static attribute data, and road environment image data, and the data extraction is performed on the traffic data to obtain multiple characteristics of the target road, including at least any two of the following: extracting data from the basic dynamic traffic data to obtain vehicle behavior characteristics of a target road; extracting data from the basic dynamic traffic data to obtain traffic flow characteristics of a target road; extracting data from the basic dynamic traffic data to obtain road state change characteristics of a target road; extracting data from the basic dynamic traffic data to obtain the road history statistical characteristics of the target road; Extracting data from the basic static attribute data to obtain road static characteristics of a target road; extracting data from the basic static attribute data to obtain road scene characteristics of a target road; and extracting the data of the road environment image data to obtain the road live multi-modal characteristics of the target road.
4. A method according to any one of claims 1 to 3, wherein the determining, based on the multi-source characteristic of the target road, a determination of the type of traffic state for the target road comprises: splicing the task conditions with the multi-source characteristics of the target road to obtain conditional fusion input information; Inputting the conditional fusion input information to a feature transformation layer to generate deep feature representation information with task adaptability; and inputting the deep characteristic representation information to a task adaptation output layer, and generating a judgment result of the target road about the traffic state type.
5. The method of claim 1, wherein the method further comprises: And in the state of receiving the abnormal signals, carrying out data extraction on the traffic data to obtain the multi-source characteristics of the target road, wherein the abnormal signals comprise any one of navigation yaw event signals, user report signals and road flow abnormal signals.
6. The method of claim 1, wherein the method further comprises: And adjusting the road state of the target road in the electronic map according to the judgment result of the target road about the traffic state type.
7. The method of claim 6, wherein the method further comprises: and feeding back route planning information to the user according to the starting point information, the end point information and the adjusted electronic map of the user.
8. A road condition detection device based on a unified model, comprising: The first obtaining module is used for extracting traffic data to obtain multi-source characteristics of a target road; The second determining module is configured to determine a result of determining, by using the multi-source feature of the target road, a traffic state type of the target road, where the traffic state type includes at least any one of a road traffic state of the target road, a road section traffic state of a road section to which the target road belongs, a congestion state of the target road, and an entrance guard state of the target road.
9. An electronic device comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, which when executed by the processor causes the electronic device to implement the method of any one of claims 1-7.
10. A computer program product comprising a computer program which, when run, causes the method of any one of claims 1-7 to be performed.

Description

Road state detection method, device, equipment and product based on unified model Technical Field The application belongs to the technical field of computers, and particularly relates to a road state detection method, device, equipment and product based on a unified model. Background In the field of navigation of the electronic map, the method has important technical value and commercial significance in real-time and accurate road traffic state excavation. Especially in high-frequency travel scenes such as network car reservation, instant freight delivery and the like, whether a road can pass is directly related to whether a driver can smoothly complete an order, whether the travel experience of passengers is smooth or not, and even the efficiency of platform operation and user satisfaction are affected. Therefore, how to quickly and comprehensively judge the road state has become a key technical problem for improving the service quality and optimizing the traffic efficiency. In the prior art, road state detection generally adopts a model architecture based on independent design of a single task. For example, a first model for judging a road closure state by analyzing vehicle track characteristics is designed, a second model for identifying a road congestion state according to real-time flow changes is designed, and a third model for judging the road congestion state by using road condition picture information is designed. The above approach to designing models on a single task independent basis suffers from the following significant drawbacks. Firstly, various models can only execute a specific single discrimination task, and a unified model architecture capable of simultaneously identifying multiple road state discrimination requirements is lacking, so that a plurality of independent models are required to be deployed in practical application to complete road state identification. And secondly, the data sources and feature dimensions adopted by the models are relatively single, and the complementary advantages of the multi-source features cannot be fully utilized, so that the recognition accuracy and the robustness of the models in complex and changeable real road scenes are obviously insufficient. Finally, the technical routes for respectively constructing the special models aiming at different discrimination tasks inevitably lead to redundancy of the system architecture, thereby not only increasing the repeated consumption of computing resources, but also obviously improving the maintenance cost and the operation complexity of the system. Disclosure of Invention The embodiment of the application provides a road state detection method, device, equipment and product based on a unified model, which can solve the technical problems of insufficient accuracy and calculation resource waste caused by the need of deploying a plurality of architectures due to the adoption of a model architecture based on single task independent design in the prior art. In a first aspect, an embodiment of the present application provides a road state detection method based on a unified model, including: carrying out data extraction on traffic data to obtain multi-source characteristics of a target road; And determining a judgment result of the target road about a traffic state type according to the multi-source characteristics of the target road, wherein the traffic state type comprises at least any one of a road traffic state of the target road, a road section traffic state of a road section to which the target road belongs, a congestion state of the target road and an access control state of the target road. In a possible implementation manner of the first aspect, the extracting data from the traffic data to obtain the multi-source feature of the target road includes: extracting traffic data to obtain various characteristics of a target road; And splicing and combining various characteristics of the target road to obtain the multi-source characteristics of the target road. In one possible implementation manner of the first aspect, the traffic data includes at least any one of basic dynamic traffic data, basic static attribute data, and road environment image data, and the data extraction is performed on the traffic data to obtain multiple characteristics of the target road, including at least any two of the following: extracting data from the basic dynamic traffic data to obtain vehicle behavior characteristics of a target road; extracting data from the basic dynamic traffic data to obtain traffic flow characteristics of a target road; extracting data from the basic dynamic traffic data to obtain road state change characteristics of a target road; extracting data from the basic dynamic traffic data to obtain the road history statistical characteristics of the target road; Extracting data from the basic static attribute data to obtain road static characteristics of a target road; extracting data from the basic static attribute da