Research Of Vehicle-mounted Road Speed-bump Defect Detection System And Method

The basic construction of urban and rural roads in our country has been completed.In order to ensure traffic safety,traffic facilities such as speed bumps have been set up on some important roads to slow down the speed.However,due to wear and manual malicious disassembly and other issues,the speed bumps have been damaged to varying degrees,which seriously endanger the safety of passing vehicles and pedestrians.At present,there is no unified detection method.In order to provide road maintenance department and drivers with abnormal defect information in time,this paper studies the vehicle-mounted road speed bump defect detection system and method based on existing road anomalies identification methods.Firstly,the detection method and technical solution are proposed.This method fuses vision and vibration information.The camera continuously captures the road in front of the vehicle and extracts the possible speed bump area.Then,when the vehicle passes by,it will generate abnormal vibration signals to filter the image recognition result.According to the proposed functional requirements,the corresponding hardware modules such as binocular vision system,acceleration sensor,navigation system are selected.According to the exitsing multi-functional road detection vehicle situation,the system will be installed on the car body to complete the detection task.Designing the computer software suitable for this system and briefly introducing the data flow.Secondly,the basic concept of convolutional neural network and the idea of attention mechanism is introduced.Designed the acceleration sensor data collection process and data enhancement method.On this basis,an improved 1D-CNN model is proposed to classify the vibration of the vehicle during normal driving or passing through speed bumps.Using different performance indicators to evaluate the model,and the relevant parameters are optimized.According to the binocular vision model and Zhang Zhengyou’s camera calibration technique,the Opencv is used to calibrate the selected camera’s parameters.The bilateral filter with better effect is selected to blur the original image,and AINDANE is used to enhance it.This paper proposes a lane detection method to extract road ROI,and integrates LSD line detection algorithm into this system for speed bump recognition.A new method to calculate the width of speed bump block is proposed for segmentation,and the statistic of Mann Kendall method is changed to make it suitable for defect detection.The paper proposed a multi-sensor decision-making fusion method which takes results of binocular distance measurement as a time bridge and vibration signal recognition as a test method for image recognition.The results show that it can effectively solve the problem of image misrecognition.Finally,three experimental schemes are proposed to test the classification effectiveness of acceleration sensor,the accuracy of speed bump recognition and binocular ranging,the integrity of system function and the real-time of transmission.The experimental results show that the improved 1D-CNN network has higher classification accuracy than other methods,with an accuracy rate of 96.27%.The error of binocular ranging is low,and the recognition accuracy can be effectively improved through multi-sensor fusion at the decision level.The system is functionally complete,the data acquisition rate meets the real-time requirements.The system and method proposed in this paper have reference value for the detection of road speed bumps or other anomalies.