Improved G-Yolov4 mobile robot target detection method

Abstract

The invention discloses an improved G-Yov4 mobile robot target detection method, which is characterized by comprising the following steps of: (1) carrying out graying preprocessing on an image acquired by a high-definition camera on a mobile robot and sending the image into a G-Yov4 network; (2) carrying out Ghost Bottlenecks operation for many times to obtain effective feature layers A1 and A2; (3) carrying out feature fusion to obtain effective features B1 and B2; and (4) carrying out Concat channel number increase once on the feature B1 and carrying out 1 * 1 convolution for five times to obtain a feature C1, carrying out downsampling stacking pooling on the obtained feature B2 and the feature C1 to obtain a feature C2, and finally outputting the effective features C1 and C2 as Yolo head. The objective of the invention is to reduce network parameter quantity and required computing power while ensuring high detection precision, optimize long operation time and more occupied resources of a current mainstream depth target detection algorithm, improve real-time target detection frame rate, and be better applied to mobile robot target detection.

Description

technical field [0001] The invention belongs to the technical field of target detection, and in particular relates to an improved G-Yolov4 mobile robot target detection method. Background technique [0002] In recent years, target detection technology has developed rapidly, and target detection algorithms such as pedestrians, vehicles, signal lights, and lane lines have also become key research directions for mobile robots. The characteristics of large size and higher computing power make it difficult to complete the deployment of embedded weak platforms, and the real-time detection frame rate is low. However, if the network is too lightweight, the occlusion of the target in the narrow space may cause problems such as false detection, multiple noise points, and missed detection of too dark targets, resulting in low detection accuracy. In response to these problems, this paper proposes an improved G-Yolov4 mobile robot target detection method, which balances the algorithm de...

AI Technical Summary

Problems solved by technology

[0002] In recent years, target detection technology has developed rapidly, and target detection algorithms for mobile robots such as pedestrians, vehicles, signal lights, and lane lines have also become key research directions. Large size and high demand for computing power make it difficult to complete the deployment of embedded weak platforms, and the real-time detection frame rate is low

However, if an overly lightweight network targets occlusions in a small space, problems such as false detection, multiple noise points, and missed detection of dark targets may occur, resulting in low detection accuracy.

Images