Research Of Object Visual Tracking For Autonomous Mobile Robot

As an important carrier of service transportation,autonomous mobile robot(AMR)plays a key role in the process of automatic processing and production,which has penetrated into the logistics and transportation,automotive electronics,airports and terminals and other industries.The complexity of application scenarios directly affects the flexibility and accuracy of AMR operation.In order to improve its adaptability to changing scenes,get rid of the dependence of AMR on industrial scenes and reduce the labor intensity of workers,the visual tracking AMR is studied,which is of great theoretical significance and application value.The main contents of this paper are as follows:Aiming at the low field of view of AMR and the over fitting and degradation of traditional convolutional neural network,the smoothing filter and histogram equalization method are used to eliminate the interference of field of view noise and improve the contrast of image objects.Based on the multi feature ratio SSD(Single Shot MultiBox Detector)algorithm,a high resolution and multi-objective clustering SER-SSD object detection algorithm is proposed,which combines residual connection and channel attention mechanism.The traditional SSD and the improved SER-SSD are tested respectively for the object data.The results show that the detection accuracy of the improved SER-SSD algorithm is improved by 7.6%and the effective confidence is more than 90%,which meets the accuracy requirements of this paper.Aiming at the problem of moving fast and easily disturbed by background for AMR object,the RGB color space of object image is transformed into Lab space.Based on fast detection KCF(Kernel Correlation Filter)algorithm and multi-scale SA(Scale Adaption)algorithm,a Sa-KCF object tracking algorithm is proposed by fusing scale adaptive method and tracking re-detection strategy.The three algorithms are tested on OTB-2013 dataset.The results show that the tracking accuracy and success rate of the improved Sa-KCF algorithm are improved by 11.8%and 29.8%respectively and the tracking efficiency is 56.31 frames per second,which meets the real-time requirements of this paper.Based on the application requirements of AMR visual tracking system,combined with the mechanical structure and basic movement mode of differential AMR,the AMR software and hardware system,including visual perception module and motion control module,is built and the AMR single object detection selection and tracking control scheme is formulated.The visual tracking effect of AMR is verified by experiments.The results show that the position of the center point of the object is maintained within±100 pixels of the center of the field of vision and the relative scale of the front and back is maintained between 80%and 120%in the tracking process.It verifies that the AMR visual object tracking system can maintain stable tracking accuracy and tracking efficiency.