Design And Implementation Of Intelligent Garbage Sorting Vehicle Based On Computer Vision

Due to the rapid growth of the social economy,people’s spiritual and cultural level has been continuously improved,and their awareness of the protection of the environment has been constantly increased.Facing the increasing domestic garbage,it is urgent to carry out garbage classification from the source,reduce the number of garbage and improve the utilization rate of garbage.With the rapid development of mechanization and intelligence,the robot technology industry continues to flourish,and the use of robots to achieve garbage sorting has broad market prospects.With the further development of computer vision technology,it is widely used in object recognition,automatic driving,target detection and other fields.The garbage sorting technology combining computer vision technology with robot technology has become a new development trend.At present,the existing garbage sorting robots in China are mainly located in the garbage yard and garbage workstation,with fixed position,high cost,and garbage sorting is carried out at the end of garbage classification.Aiming at the problem of garbage sorting,this paper designs a flexible and lightweight intelligent garbage sorting vehicle,which mainly sorts the domestic garbage generated in small areas,and strives to achieve garbage classification from the source of garbage and improve the efficiency of garbage conversion.The main thesis has the following headlines:Clear overall design objectives,select the appropriate hardware for assembly.By comparing a variety of target detection algorithms,YOLO v5 with strong applicability and small size is finally selected as the baseline network of this thesis.The data set used in this thesis is Huawei garbage classification data set.Through data analysis,the number of garbage classification data sets with unbalanced categories is balanced by rotating,changing brightness,noise blurring and adding homemade garbage classification data sets.The YOLO v5 s network model is improved.By comparing the performance evaluation results of different network models,Shuffle Net V2 is selected as the backbone network,and the network model with h-swish activation function and NAM attention mechanism is selected as the final garbage classification and recognition model.Map @ 0.5 can reach 92.1 %,and the loss function is 0.0411.Compared with the original YOLO v5 s network model,map @ 0.5is increased by 1.8 %,Precision is increased by 0.4 %,and Recall is increased by 3.7 %.The loss was reduced by 0.99 %.The model was transplanted into Jetson Nano and accelerated by Tensor RT.The reasoning speed was 21.82 ms.To ensure accuracy in calibration,three camera calibration methods are used to obtain the camera parameters.Four hand-eye calibration algorithms are used to determine the conversion matrix between the end of the manipulator and the camera,and Move It! is used to understand the motion planning of the multiaxis manipulator.The velocity angular is obtained by gyroscope,and the angle is calculated by Kalman filtering algorithm to obtain the pose state of the car.The incremental closed-loop control PID algorithm,it is used to control the movement of a car to control the movement,and the serial communication between ROS master control and STM32 is realized,so that the car searches for garbage in the specified area and realizes the garbage sorting function.