Deep Learning Method Of Vehicle Bottom Dangerous Objects Detection On Crossing Security Robot

The problem of carrying dangerous goods in cars is common and has become one of the important hazards in the public domain.Especially the location of the bottom of the car,because of its strong concealment,is easily used by criminals to hide bombs,guns,flammable and explosive dangerous goods,etc.It poses a great threat to people’s lives and property.Currently,the vehicle bottom detection mainly relies on image technology.Firstly,the vehicle bottom scanning equipment is used to capture the vehicle bottom images,and then security detection are performed manually.On the one hand,this detection method has high labor costs.Security detection need to look at the screen for a long time,which may cause false detection of eye fatigue.On the other hand,there is a single channel for obtaining information on the bottom of the vehicle,which can only use the image information to visualize the dangerous situation on the bottom of the vehicle.This detection method are not comprehensive enough.Therefore,it is extremely important to develop a fast,accurate and intelligent vehicle bottom security detection system.In recent years,the technology of intelligent security robots has developed rapidly,but it has rarely been applied in the field of vehicle bottom detection.In this paper,the existing security robot technology is used,and a set of security robot vehicle bottom detection platform is researched for the vehicle detection scene,and the algorithm of dangerous objects detection on the vehicle is improved.First,the overall system of the security robot vehicle bottom detection platform is designed in this article,including the design of the vehicle bottom detection process,the hardware and software design of the vehicle bottom detection platform.The hardware mainly includes the selection and layout design of sensors,the design of the installation and layout of the vehicle bottom detection platform;the software mainly includes the development of the vehicle bottom detection client program,the design of multi-sensor information fusion algorithm design,and the vehicle bottom dangerous goods image detection algorithm design.Through the combination of software and hardware,the vehicle bottom detection platform is highly efficient,comprehensive and intelligent.Secondly,this article focuses on the deep learning image object detection algorithm.Through systematic research on the structure of convolutional neural networks,and comprehensive comparative analysis of the current mainstream object detection technologies,the YOLOv3 network was finally selected for vehicle bottom dangerous objects detection.This paper also analyzes the feasibility and existing problems of YOLOv3 network in the detection scene of dangerous objects under the car through comparative experiments.Finally,this article introduces the establishment of the dangerous objects data set,and improves the YOLOv3 algorithm to make it more suitable for the vehicle bottom detection environment by multi-data set step-by-step training and increasing the network output branch.Verify the effectiveness of the improved algorithm.The experiment proves that the improved YOLOv3 algorithm in this paper has significantly improved the detection effect in the application of the vehicle bottom environment.The entire security robot vehicle bottom detection platform can quickly and accurately recommend the dangerous level of the vehicle bottom dangerous objects,which has great application value.