Vehicle running speed estimation method based on different scene multi-model combination

Abstract

The invention discloses a vehicle running speed estimation method based on different scene multi-model combination, which comprises the following steps of: constructing a plurality of vehicle speed estimation models for a target vehicle based on multi-scene multi-model combination in different modes; selecting a plurality of built vehicle speed estimation models as sub-models of the combined vehicle speed estimation model, and fusing the sub-models based on least square weighted fusion; dividing the different scenes into a plurality of subdivided scenes according to the obtained historical data based on the influence composition factors and the characteristic changes of the vehicle speed estimation in the different scenes; and in different scenes, adjusting the combined vehicle speed estimation model by using the weight change of each sub-model to obtain a final combined vehicle speed estimation model, and outputting a vehicle speed estimation result of the target vehicle. According to the method, the vehicle speed estimation accuracy can be improved, and different vehicle speed estimation information is integrated by utilizing a multi-model combination method, so that the purpose of improving the vehicle speed estimation precision effect is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle speed, and in particular to a vehicle speed estimation method based on a combination of multiple models in different scenarios. Background Art

[0002] Vehicle speed is one of the indicators that characterize the vehicle's operating status. For driverless cars, understanding the speed distribution types and characteristics of surrounding vehicles is a prerequisite for self-driving car behavior decisions, thereby providing safety guarantees for road driving.

[0003] There are two main methods for vehicle speed estimation: one is the global positioning system (GPS) data provided by the target vehicle to the outside. However, conventional GPS has low accuracy, positioning signals are greatly affected by environmental factors, signal loss often occurs, and reliability and stability are difficult to meet requirements. The other is vehicle speed estimation based on various external sensors, including vehicle speed estimation based on lidar point clouds and vehicle speed estimation based on camera images, but there are problems such as low measurement accuracy, long update cycle and uncontrollable delay.

[0004] Due to the complexity of various systems, the instability of the operating environment, incomplete knowledge and data islands, there are often many uncertainties in the system, which cannot be solved well by a single model. Taking the speed estimation method based on camera images as an example, due to the installation angle of the camera, when the traffic density is large, the video screen of the on-board camera is prone to vehicle occlusion. When there is a delay in transmission, the speed estimation based on the camera image cannot accurately estimate the speed, which easily leads to the ID-Switch problem in the target tracking process. Therefore, there is an inevitable uncertainty in the speed estimation method of the target vehicle by a single method. In the field of information science, information fusion technology is a better solution. Since the information used by a single speed estimation method is not exactly the same, in order to improve the accuracy of speed estimation, a multi-model combination method can be used to combine different speed estimation information, thereby achieving the purpose of improving the speed estimation accuracy. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art. To achieve the above purpose, a vehicle speed estimation method based on a combination of multiple models in different scenarios is adopted to solve the problems raised in the above background technology.

[0006] A vehicle speed estimation method based on a combination of multiple models in different scenarios includes the following steps:

[0007] Step S1, constructing several vehicle speed estimation models for the target vehicle based on multi-scenario and multi-model combinations in different ways;

[0008] Step S2, selecting several constructed vehicle speed estimation models as sub-models of the combined vehicle speed estimation model, and then fusing the sub-models based on least squares weighted fusion;

[0009] Step S3: based on the influencing factors and characteristic changes of vehicle speed estimation in different scenarios, and according to the acquired historical data, the different scenarios are divided into a plurality of subdivided scenarios;

[0010] Step S4: between different scenes, the combined vehicle speed estimation model is adjusted by using the weight changes of each sub-model to obtain a final combined vehicle speed estimation model, and the vehicle speed estimation result of the target vehicle is output.

[0011] As a further technical solution of the present invention: the specific steps in step S1 include:

[0012] A variety of vehicle speed estimation models are adopted for the same target vehicle. On the basis of maximum information utilization, several vehicle speed estimation models are constructed according to the combinatorial calculation theory. The vehicle speed estimation models include a vehicle speed estimation model based on GPS, a vehicle speed estimation model based on lidar, and a vehicle speed estimation model based on image frame difference.

[0013] As a further technical solution of the present invention: the specific steps in step S2 include:

[0014] Several models with better speed estimation effects are selected as sub-models of the combined speed estimation model, and then multi-model combined speed estimation is performed through least squares weighted fusion.

[0015] As a further technical solution of the present invention: the specific steps in step S3 include:

[0016] Based on the composition and characteristic changes of the factors affecting vehicle speed estimation in different scenarios, the target vehicle's operating scenarios are divided into multiple sub-scenarios in the form of prior knowledge according to the acquired historical data and expert experience;

[0017] By dividing the target vehicle's operating scenarios, the relevant parameters of the combined vehicle speed estimation model are automatically adjusted according to changes in scenario characteristics.

[0018] As a further technical solution of the present invention: the specific steps in step S4 include:

[0019] Between different scenarios, the combined vehicle speed assessment model is adjusted by changing the weights of each sub-model, so that the vehicle speed assessment model of each scenario is closer to the actual vehicle speed of the scenario;

[0020] Among them, the sub-model weight adjustment and scene division are determined based on a mixture of prior knowledge and vehicle speed assessment errors.

[0021] As a further technical solution of the present invention: according to the image captured by the vehicle-mounted camera, the speed of the target vehicle is estimated by a frame difference method, and a region of interest of 50m*50m is determined in the image captured by the vehicle-mounted camera;

[0022] When a vehicle enters the set region of interest ROI, the image perception algorithm outputs a vehicle target detection frame with an ID, keeps tracking it, and automatically saves the center coordinates (Xt, Yt) of the vehicle detection frame of the same tracking object;

[0023] During the entire tracking period T = [f1, f2], the displacement of the tracked object is:

[0024]

[0025] Where, L P is the displacement of the target vehicle; f1 is the frame position where the target vehicle starts tracking; f2 is the frame position where the target vehicle ends tracking;

[0026] In the two types of coordinates, the world coordinate system and the image coordinate system, the displacement length L W and displacement length L P The relationship is:

[0027]

[0028] Where D is the vertical distance between the vehicle camera and the region of interest ROI; α is the angle between the vehicle camera hardware sensing axis and the vehicle's driving direction; f c is the focal length of the camera;

[0029] Among them, f c It is obtained by the relationship between the length of the confirmed region of interest ROI in the world coordinate system and the length in the image coordinate system;

[0030] Get the average speed of the target vehicle, the formula is:

[0031]

[0032] Where V is the average speed; F is the frame rate of the on-board camera; and 3.6 is the coefficient for converting the speed in m / s to km / h.

[0033] Compared with the prior art, the present invention has the following technical effects:

[0034] The above technical solution is adopted, and the accuracy and adaptability of vehicle speed estimation are significantly improved by adopting multi-model combined vehicle speed estimation technology and combining multi-scenario adaptability. By applying multiple vehicle speed estimation models at the same time, not only the unique information contained in each model is fully utilized, but also based on the theory of combined calculation, the maximum fusion of information is achieved, thereby obtaining a more accurate calculation result. In addition, for vehicle speed estimation in different scenarios, this technology automatically adjusts the relevant parameters of the vehicle speed calculation model, making the vehicle speed evaluation model closer to the actual scene, further improving the accuracy of vehicle speed estimation. The weight adjustment of the sub-model and the scene division are determined by the hybrid of expert prior knowledge and vehicle speed evaluation error, ensuring the effective switching and adjustment of the vehicle speed evaluation model between different scenes. At the same time, the use of methods such as dynamic link library calls ensures the calculation speed of the system, providing strong support for real-time vehicle speed estimation. This multi-model, multi-scenario vehicle speed estimation technology provides a reliable data basis for applications such as vehicle control and safety warning, and effectively improves driving safety and comfort. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings:

[0036] Figure 1 A schematic diagram of the steps of a vehicle speed estimation method according to an embodiment disclosed in the present application;

[0037] Figure 2 This is a structural block diagram of a vehicle speed estimation method according to an embodiment disclosed in the present application. DETAILED DESCRIPTION

[0038] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] Please refer to Figure 1 and Figure 2 In an embodiment of the present invention, a vehicle speed estimation method based on a combination of multiple models in different scenarios includes the following steps:

[0040] Step S1: constructing several vehicle speed estimation models for the target vehicle based on multi-scenario and multi-model combinations in different ways. The specific steps include:

[0041] A variety of vehicle speed estimation models are adopted for the same target vehicle. On the basis of maximum information utilization, several vehicle speed estimation models are constructed according to the combinatorial calculation theory. The vehicle speed estimation models include a vehicle speed estimation model based on GPS, a vehicle speed estimation model based on lidar, and a vehicle speed estimation model based on image frame difference.

[0042] Step S2: Selecting several constructed vehicle speed estimation models as sub-models of the combined vehicle speed estimation model, and then fusing the sub-models based on least squares weighted fusion. The specific steps include:

[0043] Several models with better speed estimation effects are selected as sub-models of the combined speed estimation model, and then multi-model combined speed estimation is performed through least squares weighted fusion.

[0044] Step S3, based on the influencing factors and characteristic changes of vehicle speed estimation in different scenarios, and according to the acquired historical data, the different scenarios are divided into multiple sub-scenarios, and the specific steps include:

[0045] Based on the composition and characteristic changes of the factors affecting vehicle speed estimation in different scenarios, the target vehicle's operating scenarios are divided into multiple sub-scenarios in the form of prior knowledge according to the acquired historical data and expert experience;

[0046] By dividing the target vehicle's operating scenarios, the relevant parameters of the combined vehicle speed estimation model are automatically adjusted according to changes in scenario characteristics.

[0047] Step S4: between different scenes, the combined vehicle speed estimation model is adjusted by using the weight changes of each sub-model to obtain a final combined vehicle speed estimation model, and the vehicle speed estimation result of the target vehicle is output. The specific steps include:

[0048] Between different scenarios, the combined vehicle speed assessment model is adjusted by changing the weights of each sub-model, so that the vehicle speed assessment model of each scenario is closer to the actual vehicle speed of the scenario;

[0049] Among them, the sub-model weight adjustment and scene division are determined based on a mixture of prior knowledge and vehicle speed assessment errors.

[0050] In this embodiment, the speed of the target vehicle is estimated by the frame difference method based on the image captured by the vehicle camera. A 50m*50m area of ​​interest is determined in the image captured by the vehicle camera;

[0051] When a vehicle enters the set region of interest ROI, the image perception algorithm outputs a vehicle target detection frame with an ID and continues to track it. The center coordinates (Xt, Yt) of the vehicle detection frame of the same tracking object are automatically saved;

[0052] During the entire tracking period T = [f1, f2], the displacement of the tracked object is:

[0053]

[0054] Where, L P is the displacement of the target vehicle; f1 is the frame position where the target vehicle starts tracking; f2 is the frame position where the target vehicle ends tracking;

[0055] In the two types of coordinates, the world coordinate system and the image coordinate system, the displacement length L W and displacement length L P The relationship is:

[0056]

[0057] Where D is the vertical distance between the vehicle camera and the region of interest ROI; α is the angle between the vehicle camera hardware sensing axis and the vehicle's driving direction; f c is the focal length of the camera;

[0058] Among them, f c It is obtained by the relationship between the length of the confirmed region of interest ROI in the world coordinate system and the length in the image coordinate system;

[0059] Get the average speed of the target vehicle, the formula is:

[0060]

[0061] Where V is the average speed; F is the frame rate of the on-board camera; and 3.6 is the coefficient for converting the speed in m / s to km / h.

[0062] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents, and all should be included within the scope of protection of the present invention.

Claims

1. A vehicle speed estimation method based on a combination of multiple models in different scenarios, characterized in that: The following steps are involved: Step S1, constructing several vehicle speed estimation models for the target vehicle based on multi-scenario and multi-model combinations in different ways; Step S2, selecting several constructed vehicle speed estimation models as sub-models of the combined vehicle speed estimation model, and then fusing the sub-models based on least squares weighted fusion; Step S3: based on the influencing factors and characteristic changes of vehicle speed estimation in different scenarios, and according to the acquired historical data, the different scenarios are divided into a plurality of subdivided scenarios; Step S4: between different scenes, the combined vehicle speed estimation model is adjusted by using the weight changes of each sub-model to obtain a final combined vehicle speed estimation model, and the vehicle speed estimation result of the target vehicle is output.

2. The vehicle speed estimation method based on the combination of multiple models in different scenarios according to claim 1 is characterized in that: The specific steps in step S1 include: A variety of vehicle speed estimation models are adopted for the same target vehicle. On the basis of maximum information utilization, several vehicle speed estimation models are constructed according to the combinatorial calculation theory. The vehicle speed estimation models include a vehicle speed estimation model based on GPS, a vehicle speed estimation model based on lidar, and a vehicle speed estimation model based on image frame difference.

3. The vehicle speed estimation method based on the combination of multiple models in different scenarios according to claim 1 is characterized in that: The specific steps in step S2 include: Several models with better speed estimation effects are selected as sub-models of the combined speed estimation model, and then multi-model combined speed estimation is performed through least squares weighted fusion.

4. The vehicle speed estimation method based on the combination of multiple models in different scenarios according to claim 1 is characterized in that: The specific steps in step S3 include: Based on the composition and characteristic changes of the factors affecting vehicle speed estimation in different scenarios, the target vehicle's operating scenarios are divided into multiple sub-scenarios in the form of prior knowledge according to the acquired historical data and expert experience; By dividing the target vehicle's operating scenarios, the relevant parameters of the combined vehicle speed estimation model are automatically adjusted according to changes in scenario characteristics.

5. The vehicle speed estimation method based on the combination of multiple models in different scenarios according to claim 1 is characterized in that: The specific steps in step S4 include: Between different scenarios, the combined vehicle speed assessment model is adjusted by changing the weights of each sub-model, so that the vehicle speed assessment model of each scenario is closer to the actual vehicle speed of the scenario; Among them, the sub-model weight adjustment and scene division are determined based on a mixture of prior knowledge and vehicle speed assessment errors.

6. According to claim 2, a vehicle speed estimation method based on a combination of multiple models in different scenarios is characterized in that: According to the images collected by the vehicle camera, the speed of the target vehicle is estimated by the frame difference method, and a 50m*50m area of ​​interest is determined in the images collected by the vehicle camera; When a vehicle enters the set region of interest ROI, the image perception algorithm outputs a vehicle target detection frame with an ID, keeps tracking it, and automatically saves the center coordinates (Xt, Yt) of the vehicle detection frame of the same tracking object; During the entire tracking period T = [f1, f2], the displacement of the tracked object is: Where, L P is the displacement of the target vehicle; f1 is the frame position where the target vehicle starts tracking; f2 is the frame position where the target vehicle ends tracking; In the two types of coordinates, the world coordinate system and the image coordinate system, the displacement length L W and displacement length L P The relationship is: Where D is the vertical distance between the vehicle camera and the region of interest ROI; α is the angle between the vehicle camera hardware sensing axis and the vehicle's driving direction; f c is the focal length of the camera; Among them, f c It is obtained by the relationship between the length of the confirmed region of interest ROI in the world coordinate system and the length in the image coordinate system; Get the average speed of the target vehicle, the formula is: Where V is the average speed; F is the frame rate of the on-board camera; and 3.6 is the coefficient for converting the speed in m / s to km / h.

Images