Static Target Recognition of Indoor Mobile Robot Based on Improved SSD Algorithm

Abstract

In the indoor application environment, in order to improve the overall efficiency of the service robot in the target detection, combined with the deep learning target detection theory, the SSD target detection algorithm and a lightweight neural network mobilenetv2 are combined to design a new target detection algorithm, which improves the target detection efficiency to a certain extent. The main improvements of this paper are as follows. Firstly, using SSD for reference, multi-scale feature extraction is carried out for the target image, and different sizes of targets are detected at different scales; Secondly, in the process of matching the actual target detection region with the algorithm prediction region, the positive and negative samples are processed properly to ensure the stability of the model; Thirdly, using the idea of mobilenet for reference, the traditional convolution is replaced by the deep separable convolution, which greatly reduces the calculation of data processing and improves the processing speed of the model while ensuring the accuracy of the model. Fourthly, in view of the time-consuming and laborious situation of manually marking data sets, a method of automatically marking data sets is proposed, which improves the efficiency of data set preparation and reduces the workload of manual marking. Through the improvement of the above algorithm, the detection speed of the system model has been greatly improved compared with SSD algorithm, and the overall recognition efficiency of indoor service robot has been improved to a certain extent.