| Nowadays,with the continuous advancement of the development of modern cities in China,the domestic transportation is developing rapidly,and the car parc is increasing day by day.According to information released by the Traffic Administration Bureau of the Ministry of Public Security,the number of motor vehicles in the country will reach372 million in 2020,the number of cars will reach 281 million,which account for more than 75% of all motor vehicles.The result is an increasingly serious traffic accident situation.According to the statistical yearbook released by the National Bureau of Statistics,there were more than 240,000 traffic accidents across the country in 2019.People pay more and more attention to the safety of automobiles.With the development of science and technology,the real-time and correct road information can be fed back to the driver through the car assisted driving system,reducing the traffic accidents by driver’s wrong operation.Now most of the newly cars have integrated part of the driving assistance system,but for some old cars there have no driving assistance system.In view of this situation,this thesis designs some algorithms that proposes a vehicle assisted driving system based on an embedded platform,which realizes some key technologies of vehicle assisted driving on the premise of lower production cost,and has high versatility.By comparing with traditional algorithms,it is concluded that the computational efficiency of this algorithm on this platform is significantly improved compared with traditional algorithms,Ensuring the need for real-time detection.The main research content of this thesis is:（1）Explain the research content and requirement analysis of this thesis,and analyze the overall design of the system.（2）Aiming at the situation that the traditional Hough detection algorithm cannot run in real time on the Raspberry Pi 4B platform,this thesis proposes an improved realtime lane line detection module.The image is preprocessed by image morphology processing method and dynamic adaptive region of interest division,and lane line contour information is obtained by Canny edge extraction.The LSD algorithm is used to detect the straight line and output the result,and correct the lane line through the designed dynamic template tracking algorithm based on correlation coefficient.Experiments show that the algorithm is more efficient than the traditional method of obtaining regions of interest and the Hough detection algorithm on this embedded platform.The overall recognition accuracy rate in different external environments is around 93%,and detection speed reaches 40 fps.（3）Aiming at the situation that the full-image LK optical flow tracking module cannot run in real time on the Raspberry Pi 4B platform,this thesis proposes an improved real-time obstacle detection function module.After preprocessing the video image,the ORB feature in the region of interest is extracted,and the motion trajectory of the feature point is tracked by the LK optical flow method to determine the position of the static obstacle and output the result.Experiments show that this module has higher computational efficiency than the full-image LK optical flow tracking module on this embedded platform.The overall recognition accuracy rate in different external environments is above 90%,and detection speed reaches 60 fps.（4）Realize the real-time recognition function of specific sounds outside the vehicle.The algorithm performs time-frequency transformation on specific sounds such as vehicle sirens and animal calls collected from the outside world,acquire fundamental frequency information of sound through image morphology processing and endpoint detection to obtain fundamental frequency information of the sound.Then the algorithm compares these fundamental frequency information sequences with the specific sound curves to realize sound detection and recognition.（5）After algorithm optimization,the detection rate of the whole system can reach30 fps,and satisfy the needs of real-time detection. |
PDF Full Text Request