Research On Moving Objects Detection And Tracking Technology Of Service Robot Based On Computer Vision

Based on the development of social science and technology,responding to the emerging social problems,such as the accelerated population aging trend,the rising labor costs,as well as the diversified application requirements,have further promoted the "machine substitution" requirements.Care-oriented service robots can alleviate the social contradiction that young people can't accompany the elderly and the shortage of nursing staff to a certain extent.In order to make robots more intelligent,pedestrian detection and tracking technology has become an indispensable function.Due to the complexity of the indoor environment and the uncertainty of pedestrian behavior,these factors make the target detection and tracking technology based on vision still face many challenges.This paper mainly designs and implements a complete vision-based service system based on the service robot applied to the nursing home,which plays an auxiliary role in realizing the life care function of the elderly and serves the people more intelligently.This paper mainly carried out research work from the following points:(1)In order to cope with the complicated and varied working environment that occurs when the service robot is tracking,and to solve the problem that the characteristics of traditional manual design are not enough to reflect the essence of the target,a pedestrian detection algorithm based on Faster R-CNN is proposed.Firstly,the Inception-Res Net-v2 network is introduced to replace the classic VGG-16 feature extraction network,which reduces the resource consumption while extracting more fine-grained features and improving the detection performance of fuzzy small-scale pedestrians.Then,RPN is optimized according to the characteristics of the pedestrian,and the applicability correction is performed on the parameters to reduce a large number of areas without targets.In addition,the classification regression network is adjusted for the two-category pedestrian detection characteristics to ensure the accuracy of the algorithm and improve the detection efficiency.(2)In order to adress the problems of traditional objecttracking algorithms,such as unable to adapt to complex pedestrian tracking changes and the low robustness and accuracy when tracking is occluded,a single target correlation filter tracking with multi-layer convolution features(LMCFT)is proposed.Firstly,the multi-dimensional convolution feature with strong representation is combined with the correlation filtering algorithm with high real-time performance to target tracking.An adaptive update strategy is proposed to accommodate the different states of the pedestrians in the tracking process.If the target is lost due to severe occlusion,the system will use the Person-RDI algorithm to re-search for pedestrians and complete continuous tracking tasks.Finally,the experimental results show that the LMCFT algorithm has strong processing ability for similar object interference,target appearance deformation and long-term occlusion,and can ensure high precision in practical environment applications.(3)For the elderly,it is easy to fall due to the inability of the legs and feet,and in order to ensure the first time to obtain rescue,a fall detection algorithm that combines multiple features of the human body is proposed.Firstly,based on the analysis of the human fall behavior,the feature with high discrimination is selected as the basis for discrimination.The human body features based on time series are extracted and sent to the circular queue.In order to ensure real-time performance,the SVM classifier is used to discriminate the fused feature vector.At the same time,considering the occlusion factor,the occlusion rate discrimination is added to the fall detection.The experimental results show that the algorithm has high accuracy and low miss detection rate,which provides an effective guarantee for the detection of human fall behavior in the actual indoor scene.