A mobile device-oriented obstacle detection method

Abstract

The invention discloses an obstacle detection method for mobile terminal equipment, comprising: (1) selecting an obstacle detection model; (2) taking the recognition accuracy a of the trained detection model as a dependent variable, the depth of the model d, the Width w and image size r are used as independent variables; (3) prune the depth d, width w and image size r of the model respectively, and fine-tune the model on the data set, and use the function f to fit a with d, w , r; (4) Solve the optimal depth, width and picture size (d) through the optimization function m , w m , r m ), and according to the obtained (d m , w m , r m ), derive the corresponding simplified model structure; (5) train the simplified model on the data set; (6) run the simplified model on the mobile device, input the surrounding environment photos in real time, and identify obstacles. The present invention enables the deep convolutional neural network to meet the fluency requirements of mobile terminal equipment, and at the same time enables obstacle detection with high accuracy.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and in particular relates to an obstacle detection method for mobile terminal equipment. Background technique [0002] With the development of computer vision technology, technologies such as image recognition and object detection are widely used in areas such as automatic driving and access control systems. The deep convolutional neural network is one of the most important deep learning frameworks in the field of computer vision. It consists of multiple convolutional layers and fully connected layers. Currently, the most accurate image recognition, target detection and other algorithms use deep convolutional neural networks. product neural network. However, the operation of deep convolutional neural networks consumes a large amount of computing resources, and most mobile devices (such as mobile phones, on-board chips, etc.) have very limited computing resources, which seriously hinders ...

AI Technical Summary

Problems solved by technology

However, in actual use, it is found that given a calculation amount, simply reducing the width, depth or image size will greatly reduce the accuracy of the model

On the contrary, if the size of the three can be reduced at the same time and an optimal combination strategy is found, the loss of model accuracy can be minimized, but how to find the optimal combination of the three is a difficult problem to solve

[0005] At present, the existing technology is to find the optimal combination by means of approximate brute force search, that is, to train a new model with the possible depth, width, and picture size that meet the calculation requirements, and select the one with the highest accuracy, but This method is very time consuming

Images