Detection method for flexible object in uncertain direction, electronic equipment and storage medium

Abstract

The invention discloses a method for detecting a flexible object in an indefinite direction, electronic equipment and a storage medium, and belongs to the field of computer vision. The method comprises the following steps of: 1, receiving external input image data; 2, performing detection through the trained FCN depth model; step 3, predicting input external image data by using the trained FCN depth model to obtain a detection result and outputting the detection result; and 4, carrying out training iteration on the model, and transmitting a detection error for training iteration. And finally, completing the detection of the flexible object in the uncertain direction by adopting the method of the third step and the fourth step for multiple times in sequence. The invention provides an accurate and efficient method for detecting the flexible object in the uncertain direction, and provides more accurate interpretation and analysis for the detection of the flexible object in the uncertain direction in real life.

Description

technical field [0001] The invention relates to the field of computer vision, and is particularly characterized in that a method for detecting flexible objects with indefinite directions based on an FCN depth model is proposed for the characteristics of flexible objects. Background technique [0002] It is necessary to detect flexible objects with uncertain directions in the video detection system. For example, when performing equipment maintenance inspections on the power grid, fences must be erected to limit the entry and exit of the construction site. In the surveillance video, there are many targets that need to be automatically detected for maintenance inspections. Not only do people and clothing need to be detected, but also large-span flexible targets such as seines need to be identified to divide the work area. However, the direction of the flexible object is uncertain, and there are cases where it straddles the monitoring window to a large extent. [0003] The trad...

AI Technical Summary

Problems solved by technology

In this way, it is not conducive to the subsequent training and prediction of the model

[0005] 2) In the data training stage: Considering that the training picture has a lot of environmental interference information, if the traditional rectangular label is used for flexible objects, the difficulty will increase exponentially when performing bounding box regression during training, and the corresponding detection effect indicators will be greatly reduced.

[0006] 3) In the prediction stage: Due to the accumulation of data cleaning and training errors, a detection model with good detection effect cannot be obtained, and the accuracy of the prediction results cannot reach the target

Images