Method for combining FCW system and improved YOLOX-S target detection algorithm
By combining the FCW system with the improved YOLOX-S target detection algorithm, the problems of high costs and maintenance difficulties in traditional collision warning systems are solved, and accurate and reliable early warning information is achieved, sensor costs are reduced, and system convenience is improved. Economical.
Patent Information
- Application Number
- CN202410998569.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-24
- Publication Date
- 2024-11-19
AI Technical Summary
Traditional anti-collision warning systems rely on ultrasonic or radar sensors with high cost and difficulty in maintenance, and the measurement accuracy is unstable in severe weather and have limited detection distance.
Combining the FCW system and the improved YOLOX-S object detection algorithm, it reduces sensor costs and provides accurate and reliable early warning information through image data acquisition, model training, object detection and safe distance calculation.
It realizes the provision of accurate and reliable early warning information while reducing the cost of sensor purchase and maintenance, and improves the convenience and economy of the system.
Smart Images

Figure CN118982809A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of automotive electronic technology, and in particular to a method for combining an FCW system with an improved YOLOX-S target detection algorithm. Background Art
[0002] At present, traditional anti-collision warning systems mainly use ultrasonic or radar sensors to obtain information about the vehicle's driving environment to achieve multi-directional monitoring. However, these high-precision sensors have some disadvantages, such as high cost and difficult maintenance. For example, millimeter-wave radar is widely known for its high-precision positioning ability, which can accurately locate the position of obstacles. However, the signal loss during transmission is large, resulting in its limited detection range. In contrast, lidar has a longer detection range and high-precision measurement capabilities. However, in bad weather, its measurement accuracy may be affected by large fluctuations.
[0003] With the rapid progress of deep learning-based target detection technology in the field of image detection, the detection accuracy and processing speed of computer vision have been able to be applied in engineering, which provides the possibility of applying image detection technology and measurement algorithms to the development of traditional FCW systems. By combining the relevant calculations of computer vision with the anti-collision warning system, the problem of high sensor costs and maintenance costs can be solved. Summary of the invention
[0004] The purpose of the present invention is to provide a method combining the FCW system with an improved YOLOX-S target detection algorithm, which is easy to deploy and can provide accurate and reliable early warning information, while reducing the purchase and maintenance costs of the sensor, bringing a more convenient and economical solution to users.
[0005] To achieve the above object, the present invention provides a method for combining the FCW system with an improved YOLOX-S target detection algorithm, comprising the following steps:
[0006] S1, collect image data and build the data set required for model training;
[0007] S2, resampling the collected data set;
[0008] S3, build an improved YOLOX-S algorithm model;
[0009] S4, dividing the data set obtained after the resampling process into a training set, a validation set and a test set;
[0010] S5, aggregating the resampled image data into a data set, and labeling the image data by category;
[0011] S6. Train and deploy the improved YOLOX-S algorithm model;
[0012] S7, collecting road targets and inputting the collected results into the improved YOLOX-S algorithm model for target detection;
[0013] S8, collecting vehicle speed signals and sending the vehicle speed through the CAN bus;
[0014] S9, integrating the improved YOLOX-S algorithm and the SGBM stereo matching algorithm, obtaining the vehicle speed data and calculating the safe distance between the detected target and the vehicle at the current speed;
[0015] S10, determining the safe distance between the vehicle and the target at the current vehicle speed, and triggering the FCW function if the safe distance is less than the safe distance.
[0016] Preferably, in step S1, the categories of targets to be collected are established, and image data of cars, trucks, bicycles, motorcycles and pedestrians are collected respectively.
[0017] Preferably, in step S3, the YOLOX-S target detection algorithm is selected as the baseline model, and the improvement process of the baseline model YOLOX-S algorithm is as follows: first, a convolutional block attention module is added to the CSPDarknet part of the YOLOX-S algorithm; then, the SiLU activation function originally used by the YOLOX-S algorithm is changed to a Softplus activation function.
[0018] Preferably, in step S4, 70% of the data in the data set obtained after the resampling process is selected as a training set to train the improved YOLOX-S algorithm model constructed in step S3, 10% of the data is used as a validation set, and 20% of the data is used as a test set.
[0019] Preferably, in step S5, the labelimg image annotation tool is used to annotate the image data, and the image data is respectively labeled into five categories: car, truck, bicycle, motorcycle and pedestrian.
[0020] Preferably, in step S6, the improved YOLOX-S algorithm model is deployed on a deep learning development board for training.
[0021] Preferably, in step S7, a binocular camera is used to capture the road target, and the left camera module of the binocular camera captures image information and outputs the position information between the vehicle and the target.
[0022] Preferably, in step S10, the driver is reminded to take braking action by at least one of sound, display, vehicle braking, and vibration warning.
[0023] Therefore, the present invention adopts the above-mentioned method of combining the FCW system with the improved YOLOX-S target detection algorithm, which is combined with the target detection algorithm, is easy to deploy, can provide accurate and reliable early warning information, and at the same time reduces the purchase and maintenance costs of the sensor, bringing users a more convenient and economical solution.
[0024] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1 It is a flow chart of a method of combining the FCW system with the improved YOLOX-S target detection algorithm of the present invention;
[0026] Figure 2 It is a schematic diagram of a YOLOX-S network model of a method combining an FCW system with an improved YOLOX-S target detection algorithm of the present invention;
[0027] Figure 3 It is a schematic diagram of a CBAM module of a method combining an FCW system with an improved YOLOX-S target detection algorithm of the present invention;
[0028] Figure 4 This is a comparison chart between SiLU function curve and Softplus function curve;
[0029] Figure 5 It is a flow chart of a method for combining the FCW system with the improved YOLOX-S target detection algorithm of the present invention;
[0030] Figure 6 It is a warning decision flow chart of a method of combining the FCW system with the improved YOLOX-S target detection algorithm of the present invention. DETAILED DESCRIPTION
[0031] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.
[0032] Embodiment 1
[0033] like Figure 1 As shown, the present invention provides a method for combining the FCW system with the improved YOLOX-S target detection algorithm, comprising the following steps:
[0034] S1. Collect image data and determine the categories of targets that need to be collected. Collect image data of cars, trucks, bicycles, motorcycles, and pedestrians respectively to build the data set required for model training. The above categories basically cover the targets that a driver can detect while driving.
[0035] S2. Since the collected data set may have an unbalanced category distribution problem, the collected data set is partially resampled to expand the data set size so that the proportion of each data set is close.
[0036] S3, select the YOLOX-S target detection algorithm as the baseline model, and then build an improved YOLOX-S target detection algorithm model, such as Figure 2 The improvement process of the baseline model YOLOX-S algorithm is as follows: First, a convolutional block attention module (CBAM) is added to the CSPDarknet part of the YOLOX-S algorithm, as shown in Figure 3 As shown in Figure 1, the introduction of this module can improve the perception of the model, thereby improving the detection accuracy of the improved algorithm. Then, the SiLU activation function originally used by the YOLOX-S algorithm is changed to the Softplus activation function. The model can improve its fitting performance by using the Softplus activation function, as shown in Figure 1. Figure 4 shown.
[0037] The function expression of the SiLU activation function is:
[0038]
[0039] The function expression of the Softplus activation function is:
[0040] f(x)=log e (1+e x )
[0041] S4. Divide the data set obtained by resampling in step S2, select 70% of the data as the training set to train the improved YOLOX-S target detection algorithm model constructed in step S3, select 10% of the data as the validation set, and select 20% of the data as the test set to evaluate the prediction results of the model. Such division can effectively verify the generalization ability and performance of the model on different data sets.
[0042] S5. Use the labelimg image annotation tool to classify the resampled image data in step S2. Label these image data into five categories: car, truck, bicycle, motorcycle, and pedestrian. These five categories basically cover all kinds of situations on the road that need to trigger the FCW function. By annotating the categories, accurate annotation data can be provided for model training, thereby improving the model's target detection performance and accuracy.
[0043] S6. Deploy the improved YOLOX-S model on the deep learning development board for training.
[0044] S7. Use a binocular camera to collect images of vehicles, pedestrians and other targets on the road. The left camera module of the binocular camera simultaneously collects image information and outputs the position information between the vehicle and the target. The collected results are input into the improved YOLOX-S model in step S6 for target detection.
[0045] S8, collect vehicle speed signals and send the vehicle speed through the CAN bus.
[0046] S9, integrating the improved YOLOX-S algorithm with the SGBM stereo matching algorithm, such as Figure 5 Among them, the vehicle speed data sent through the CAN bus in step S8 is obtained, and the safe distance between the detected target and the vehicle when the vehicle is traveling at the current speed is calculated.
[0047] S10: Determine the safe distance from the target at the current vehicle speed. When the calculated distance from the target at the current vehicle speed is less than the safe distance, trigger the FCW (Forward Collision Warning) function to remind the driver to take necessary braking operations to avoid accidents by at least one of sound, display, vehicle braking, and vibration warning. The warning decision process is as follows: Figure 6 shown.
[0048] Therefore, the present invention adopts the above-mentioned method of combining the FCW system with the improved YOLOX-S target detection algorithm, which is combined with the target detection algorithm, is easy to deploy, can provide accurate and reliable early warning information, and at the same time reduces the purchase and maintenance costs of the sensor, bringing users a more convenient and economical solution.
[0049] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.
Claims
1. A method combining the FCW system with the improved YOLOX-S target detection algorithm, It is characterized in that The following steps are involved: S1, collect image data and build the data set required for model training; S2, resampling the collected data set; S3, build an improved YOLOX-S algorithm model; S4, dividing the data set obtained after the resampling process into a training set, a validation set and a test set; S5, labeling the resampled image data by category; S6. Train and deploy the improved YOLOX-S algorithm model; S7, collecting road targets and inputting the collected results into the improved YOLOX-S algorithm model for target detection; S8, collecting vehicle speed signals and sending the vehicle speed through the CAN bus; S9, integrating the improved YOLOX-S algorithm and the SGBM stereo matching algorithm, obtaining the vehicle speed data and calculating the safe distance between the detected target and the vehicle at the current speed; S10, determining the safe distance between the vehicle and the target at the current vehicle speed, and triggering the FCW function if the safe distance is less than the safe distance.
2. A method of combining the FCW system with the improved YOLOX-S target detection algorithm according to claim 1, Features: In step S1, the categories of targets to be collected are determined, and image data of cars, trucks, bicycles, motorcycles and pedestrians are collected respectively.
3. A method of combining the FCW system with the improved YOLOX-S target detection algorithm according to claim 1, Features: In step S3, the YOLOX-S target detection algorithm is selected as the baseline model. The improvement process of the baseline model YOLOX-S algorithm is as follows: first, a convolutional block attention module is added to the CSPDarknet part of the YOLOX-S algorithm; then, the SiLU activation function originally used by the YOLOX-S algorithm is changed to the Softplus activation function.
4. The method of combining the FCW system with the improved YOLOX-S target detection algorithm according to claim 1, Features: In step S4, 70% of the data in the data set obtained after the resampling process is selected as a training set to train the improved YOLOX-S algorithm model constructed in step S3, 10% of the data is used as a validation set, and 20% of the data is used as a test set.
5. The method of combining the FCW system with the improved YOLOX-S target detection algorithm according to claim 1, Features: In step S5, the labelimg image annotation tool is used to annotate the image data into five categories: car, truck, bicycle, motorcycle, and pedestrian.
6. The method of combining the FCW system with the improved YOLOX-S target detection algorithm according to claim 1, Features: In step S6, the improved YOLOX-S algorithm model is deployed on the deep learning development board for training.
7. The method of combining the FCW system with the improved YOLOX-S target detection algorithm according to claim 1, Features: In step S7, a binocular camera is used to collect road targets, and the left camera module of the binocular camera collects image information and outputs the position information between the vehicle and the target.
8. The method of combining the FCW system with the improved YOLOX-S target detection algorithm according to claim 1, Features: In step S10, the driver is reminded to take braking action by at least one of sound, display, vehicle braking, and vibration warning.
Citation Information
Patent Citations
Vehicle front collision early-warning method based on vehicle-mounted binocular camera
CN106240458A
Training method, using method and training device of YOLOv5s model
CN115761404A
