Visual Tracking And Machine Learning Of Electric Car Seat Manipulator

At present,the intelligent technology is gradually integrated into the industrial production,living services and other fields.Artificial intelligence and machine learning has become one of the hot topic of attention.The perfect combination of machine learning,computer vision,data mining can solve a lot of problem of manufacturing process.Such as the electric inspection and position adjustment of the automobile electric seat before delivery.The paper intends to use machine learning and computer vision theory to realize the seats rapid adjustments and solve the problem of manipulator follow-up seat switch in the process of inspection and position adjusting.And the Software and hardware system was built based on those method.In the process of adjusting,the parameter changes of every joint parameters have an effect on result.In order to solve this coupling phenomenon,a method based on machine learning was proposed.First of all,this method analyze the characteristics of the mechanical structure of the seat,proved that the position relationship among the special surface points is linearly related,and then use laser ranging sensor to obtain the position information of specific points on the surface of electric seat,set up the segmented ridge regression model which was ridge regression model after further trained by linear classifier.The segmented ridge regression model can predict seat position and orientation accurately and eliminate the influence of abnormal points.The experiment shows that,after the judgment of the segmented ridge regression model,every switch needs to be adjusted just once to reach the standard.And the production efficiency was raised.In the process of adjusting,the switch of the seat with seat do planar motion together.In order to solve the follow-up problem of the manipulator and switch,local phase correlation(LPC)visual track algorithm is proposed.At first,this algorithm divide the original image into several regions,and train support vector machine(SVM)model use features of seats and distractions which was extracted by the improved HOG algorithm in each region of the image.Then through the model eliminate distractions region for each frame,each region,and through phase correlation method calculate the translation and rotation between two adjacent frames which was filtered,At last,the LPC result as the motion parameters passed to the manipulator control system.The experimental results show that after predict of the LPC,the instantaneous error and accumulated error of the image position variation meet the control requirements.For the realization of those algorithms,the paper has also built the software and hardware platform.First of all,according to automobile electric seat requirements of position adjustment and electrical inspection,the paper has done the overall design for automobile electric seat manipulator system,and built hardware environment,including the mechanical structure and the peripheral circuit of the manipulator.Then according to hardware requirements,the paper has written the hardware drivers for Linux,and developed seat test system software by using QT cross-platform application development framework.The software includes the realization of the algorithm,the management of product quality,control of hardware,etc.Finally after debugging,the visual track and position adjustments of manipulator for automobile electric seat have achieved the expected effect.