Method for determining target single pixel contour

Abstract

The invention discloses a method for determining a target single pixel contour. The method comprises the following steps: step 1, reading an image of a vehicle to be detected, marking the image as I1,wherein the height of the image is height and the width is width; step 2, inputting I1 into a full convolutional neural network and obtaining a contour probability graph I2; step 3, finding a pixel point with the largest pixel gray value in I2 and marking said pixel point as PL, and marking the gray value of said pixel point as PL (xL, yL), wherein (xL, yL) denotes the coordinates of the pixel point PL. The invention has the beneficial effects that: through performing single pixel contour extraction operation on a target by using the method of the invention, interference of the background adjacent to the edge of the target can be avoided, and because of the characteristics of the full convolutional network, the target contour extraction method is not affected by colors of the target and illumination, and contour location can be performed well when the color of the target is close to that of the background, and a good target contour extraction effect in a complex environment can be achieved.

Description

technical field [0001] The invention relates to the fields of computer vision and machine learning, in particular to a method for determining target single-pixel contours by combining fully convolutional neural networks and local contour maximum probability calculations. Background technique [0002] With the development of image processing technology, the contour extraction of the target object is increasingly concerned by researchers. The extraction of the target contour plays an important role in the recognition and positioning of the target object. Secondly, with the extensive development of neural networks, pixel-level The acquisition of training samples for the classified network is still a major problem, and the accurate single-pixel contour extraction method of the target contour can save the manpower and time required for frame selection of samples. [0003] At present, many scholars have proposed different target contour extraction methods, which can be mainly divi...

AI Technical Summary

Problems solved by technology

[0005] In summary, the current target contour determination methods have the following deficiencies: 1) There are many interference edges in complex backgrounds, which may easily cause difficulty in identifying the target contour; 2) The information in the vicinity of the target will interfere with the precise positioning of the target; 3) There is no Make full use of the profile's own information

Images