Method for positioning and navigation by use of picture retrieval

Abstract

The invention provides a method for positioning and navigation by use of picture retrieval. The main content of the method comprises training data sets of scenic spots, inputting an image, performing a VQA (Visual Question Answering) model, predicting probability by use of a multilayer perception classifier and inputting the probability into a navigation software to acquire route information; the method comprises the processes of: firstly, creating a database by use of Chinese famous scenic spot pictures for training; when a scenic spot position is searched, importing a scenic spot image in a search bar, determining important words and image regions by use of the VQA model including a back propagation guidance technology and a blocking model; in combination with convolutional neural network (CNN) and Long Short-term Memory (LSTM), predicting answer probability distribution by using a multilayer perceptron as a classifier; and finally inputting an answer into the navigation software to acquire the route information. The method breaks through the limit that a destination can only be found and positioned by character input before; whereas through the method, a target location can be exactly searched and positioned even if people do not know the geographical name or cannot use the local language so that people travel more conveniently.

Description

technical field [0001] The invention relates to the field of image retrieval and positioning, in particular to a method for utilizing image retrieval, positioning and navigation. Background technique [0002] Positioning and navigation technology has brought great convenience to people's travel for traditional methods, but the existing positioning and navigation technology is limited to text input for positioning and destination search, for those who do not know place names or do not know how to use In terms of the local language, there are still many inconveniences. [0003] The present invention proposes a method of using pictures to search for positioning and navigation. It adopts pictures of famous scenic spots in China to establish a database for training. When searching for the location of scenic spots, import the images of scenic spots in the search bar, and use the VQA model, including guided backpropagation technology and Occlusion model to determine important word...

AI Technical Summary

Problems solved by technology

[0002] Positioning and navigation technology has brought great convenience to people's travel for traditional methods, but the existing positioning and navigation technology is limited to text input for positioning and destination search, for those who do not know place names or do not know how to use In terms of the local language, there are still many inconveniences

Images