Design And Implementation Of Intelligent Vehicle Sensing System Based On Computer Vision

With the rapid development of the national economy,level of ascent in people 'life,car ownership raised gradually,which has led to an increase in the number of traffic safety accidents,and the number of traffic accident deaths remains high.High-level automatic driving technology cannot be applied to the market immediately to improve the driving safety of vehicles.Therefore,the advanced assisted driving system(ADAS) is currently the universal choice to help drivers drive safely.The results of foreign studies showed that the driving recorder can also restrict driving behavior and improving driving safety.No matter which electronic technology method is used to improve the driving safety of a car,it is inseparable from the perception of the surrounding environment of the driving car,so the visual perception technology has become a hot spot for various component suppliers and industry research.By analyzing the product positioning,application scenarios and research development status of the perception system,it is feasible to protect the safety of pedestrians from the perspective of warning the driver and increasing the illegal cost of pedestrians running red lights.The existing ADAS driving recorders on the market also have a driver warning function,but there is a problem of low warning accuracy.Therefore,the specific goal is to design a complete system containing hardware and software,to perceive the environment from the aspects of vision and motion posture,to realize the functions of collecting and saving data,ADAS driving reminders,uploading data to the platform,etc.Also,it is supposed to achieve low power consumption during sleep,Good real-time performance,high robustness,easy to expand and transplant performance goals.In terms of hardware,the power supply is designed as multi-channel parallel power supply to ensure stable operation of the high-power unit and easy control.The data processing board is responsible for image acquisition,visual detection algorithm calculation,file management and other tasks.The MCU processes some real-time high-density and low-density sensor data and manages the system power supply.YOLO V3 algorithm contributing to achieving better detection and identification of the key elements of the environment.Combined with GIo U,it explained the frame coincidence better,redesigned the loss function,and used cross entropy to improve the detection of small targets.The system achieved a high performance on the calculation server.In addition to improving the algorithm,the software also wrote corresponding drivers during the process of transplanting the operating system.Also,we designed a visual program based on the Qt platform,including real-time recording,file management,system settings,video playback,system information and other interfaces.That made it more efficient to view files,upload data,set function switches,etc.Finally,through the verification test,both of the hardware and software of this system showed a high performance achieving all the preconceived functions.In the working mode,the error of each power supply voltage is less than 50 mV,the sleep mode system power is about 1.2mW,the data transmission delay is less than 2s when the 4G signal strength is greater than 30%,and the Recall recognition for key elements of the driving environment reaches 81.38%,and the processing speed can be stabilized 30 fps.In the future,binocular cameras will be used to improve the shortcomings of distance insensitivity,target tracking algorithms will be added,and the types of key elements for detection and identification will be increased to make the system more complete.