Target Detection Using Multi-sensor Fusion On A Robot

With the development of computer technology and artificial intelligence,more and more applications use mobile robots to perform tasks in dangerous environments or where human beings are difficult to reach.The mobile robot can autonomously make decisions by itself and performing a pre-set task upon reaching the target location by its sensors and processors.The mobile robot has the ability to make decisions,judge,and execute predefined tasks after reaching the destination.However,due to the uncertainty of the unknown environment,single sensor is difficult to satisfy the requirements of complex tasks,so multi-sensor data fusion technology could integrate the information of each sensor,and then get more comprehensive and accurate decision information than single sensor.We designed and developed a target designation system by using the idea of multi-sensor data fusion in Khepera IV embedded robot.In unknown environment,the robot uses the built-in camera and ultrasonic sensor to move,avoid obstacles and find the target by the built-in camera and ultrasonic sensor.First of all,the robot uses built-in camera to shoot images from the different angles.After that,robot extracts the characteristics of the target in the process of image preprocessing,and store into the host computer.In the process of movement,robot detect the surrounding objects to discover target using ultrasonic sensors,and uploading the image of objects to the host computer to pick out the target.The research content includes the following several aspects:1.Using zabbix open source software,a robot monitoring system is built in the virtual machine of the Windows host computer.Real-time information of the robot is monitored,such as power usage,CPU utilization and motion trail of the robot.The information is stored in the MySQL database in order to be used in the process of motion control and data fusion.In addition,data changes can be intuitively shown in the web-based data interactive pages of zabbix software.2.Modeling motion of the robot and combined with the hardware configuration of Khepera IV robot,the speed control,position control and direction control methods are analyzed.A fuzzy obstacle avoidance strategy is put forward,which is used in the process of robot movement.By simulation,the fuzzy obstacle avoidance strategy is effective.3.A target recognition method base on machine learning and image processing is presented in this paper.Firstly,the characteristics of the target feature are extracted by image preprocessing.Before the target identification,the images of the target were taken by robot at different distances and perspectives,and 200 images containing the target were used as positive samples,300 images without the target were used as negative samples.Using the HOG feature extraction method to extract image gradient feature,the positive and negative samples are marked.Training these samples by linear SVM classifier training,a two class classifier was gotten.And then,robot shoots around the object in the process search of target objects,and uploading to the host computer for image processing through image enhancement,thresholding,edge detection,Hof transform and others pre-processing to divided into the target area from picture.Identify this area with SVM classifier to judge whether to determine whether the target,and estimate the location of the target.4.In robot target calibration,we build the robot monitoring system,accomplish data transmission and real-time image processing.The experimental result manifest that the robot can complete the target recognition task in the unknown environment,and prove the effectiveness of the system designed.