IROS2020一种激光SLAM算法原创

Hi I'JA乐Le I'm pleased to present our work itriinic and temporal calibration of automotive freedom and three d lightar which is down together with宇寒X洁制王and文N杰林。Let's start from demonstrating videos to show the importance of calibration LA can providestrated structure information of the environments which is shown as y point on the other hand radar only provides two d geometical information of ours which are shown as greenfhers。In the video we can notice thatt extreme calibration must be done such that radar measurements and ladar measurements can be well aligned and useful further applications。However when the motion of the censor is getting larger especially when the vehicle is changing its direction a mismach between radar and LA measurements would occur despite good extenic calibration this video is recorded from new scene data set we can observe that while the vehicle is making a turn the light measurements are running after the radar measurement it is more obvious at the edge of the role。

Hence temper calibration is critical and necessary here we go through the overview of calibration procedures in this work strength calibration and temper calibration are accomplishedequently。And we collected to data set for each of them。In the beginning the target center as the me processes thetric data set and outputs the target center as the。By establishing the correspondences between the target centers from radar and lightdar seeto erinic parameters are estimated afterward temper calibration follows the same procedure to obtain target centers given thetrinsic parameters the t delay can be estimated by special temple relationship。

So far only a few works address calibration between radars and other position sensors one of the most challenging issue is that radar only offers two dimensional geometric measurements but the rigid transformation between the coin systems of radar and other three d censers must be estimated。To solve this issue proceed proposed to use radar echo intensities in other words r CS values to infer the missing dimension of the p radar since r CS measurement has different value distribution across radars ver field of view。We follow their utimization procedures and Turkey design which is easily detected by radar and lightdar。

In our settings thetic data set is collected while the vehicle is driven slowly to reduce the influence of the t delay。Therefore we estimate a exre parameters after target center estimation and correspondence establishment。However RA can provide accurate language range measurement one problem could occur when we try to estimate the target center from the LA measurement。In this figure we can observe that the target measure become spas when the target is located at long ranges resulting in the decrease of the accuracy of Turkey detention。Instead of detecting targets directly we adopt a mapping localization approach to perform target center estimation。First would be a map of the experiment environment and mark the target centers in the map。

To estimate the tu center position in the point called map。Triangle model fifty method is adopted。We extra the edge points of theangle board from the denal map and use them to feed a triangle modelto。Afterward we can estimate the location of the target for each light skin by performing localization。In the video the Turkey is placed in front of the senss why points are the Turkey measurements from lyar west fear is the estimated target center。We can notice even the target measurements become spas as the vehicle is gradually driven away from the target we can still accurately estimated target centers by mapping organization approach after the target centers are estimated the radar light correspondences can be established fortinic and temporal calibration。

However given a goodten calibration the t delay would still cause a mismatch between radar and ladar measurement when the motion of sensors is large。To solve this issue temper calibration between radar and ladar is necessary。We propose a target based approach for temple calibration in this work we assume thei delay is a constant and estimate tie delay using special temple relationships of the target。Since t delay is a one dimensional parameter sufficient features in amas angle would be enough for the utimization we utimize the tiny delay to minimize the umas error which is DeFined as the difference between the uma angle of radar tuck detens and the corresponding lightder Turkey estimation。

If perfectenss exist the t delay could be estimated by arbitra emotion however the t delay estimateation becomes uncertain due to the censor uncertainty。To reduce the influence from uncerta and ensure the observability of t delay rappe motion is preferred in the data collision procedure。In addition asimulation is performed according to the sensor sub specification and experiment set up the result shows that the tight delay estimation error decreases with rapp motion which support our belief。We propose To Generate motion by moving the sensor system include radar and lightar where the targets are stationary。

This video demonstrates how we replace the targets and generate the motion with the sensor system in practice instead of given a translational motion we reta a censor system which generates motion more easily and flely。To verify the performance of temper calibration we compsate radar and light measurements into the same frame。In this video the red and the green points represent the Turkey de detection from radar before an after temperal calibration。It is obvious that the Turkey detentions from radar are more close to the Turkey estimation from radar after temple CA。The effectiveness of temper calibration is confirmed。In summary we have To Contributions。

First represent a map localation method to estimate target location which improves the accuracy of target center estimation and long range from the lighter measurements。Second we also propose a novel temple cation procedure to estimate the t delay between radar and lightar。The required motion for temper calibration is also analyzed。For discussion it should be noted that uncertainties in the map and the localization module could downgrade the performance of Turkey center estimation。The accuracy of the map and localization approaches should be analyze and could be further improved to obtain better calibration results。Besides sufficient motion during data collection are required for temple calibration but motions could maketen calibration less reliable。

Since data requirements for extrinsic and tempor television are different they are performed separately in this work。Further improvements on utimization and integ of both calibration tasks will be our future work。It is the end of our presentation if you have any questions with work please feel free to contact us。Thanks for your watching。