Method for colorizing vehicle-mounted infrared video based on contour tracing

Abstract

The invention provides a method for colorizing a vehicle-mounted infrared video quickly, which obtains contour regions of object categories in each frame by tracing contour feature points, and transfers colors to each region by category feature colors. The method comprises the steps of: first constructing a feature color collection of scenery samples by the feature color of each scenery category represented in a natural color image; next clustering infrared key frames by an improved and effective method of K-Means to obtain segmenting regions and to extract contour feature points; then tracing the feature points through a KLT algorithm to obtain locations of the feature points in the next frame and to correct at the same time, and restoring the contour by adopting a technique of B-spline interpolation to obtain the contour regions of each category of the frame; and finally endowing the distinguished category regions of each frame with the feature colors according to the categories, so that the object is colored by suitable colors and a quick colorization of infrared video sequence is realized. Experimental results show that the method of the invention improves the processing speed and can receive a satisfactory result.

Description

technical field [0001] The invention relates to a fast colorization method for vehicle-mounted infrared video. The method can endow the vehicle-mounted infrared image with more natural color, improve the processing speed, and realize rapid colorization of the vehicle-mounted infrared video. Background technique [0002] The on-board night vision system can assist driving at night and improve driving safety. On the night vision system screen in the cab, night vision images of road conditions and roadside trees and buildings that cannot be obtained by human eyes or ordinary visible light camera equipment can be displayed. However, because the night vision images obtained at night are blurred, and the targets monitored or detected by night vision devices are far away, it is often difficult to make quick and accurate judgments on the content of night vision images as in daytime natural light images. . Since the color level that the human eye can distinguish is hundreds of time...

AI Technical Summary

Problems solved by technology

[0003] At present, the field of color night vision mainly studies the colorization of static night vision images. In order to achieve a better color image effect, the algorithm is often designed with a high time complexity. This type of algorithm cannot realize the colorization of dynamic night vision images in real time.

However, there are relatively few real-time algorithms for dynamic night vision colorization. Related literatures directly use the method of extracting motion vectors from video coding for colorization. Since motion vectors are mainly used for video compression and cannot reflect the real sports field, there are often Error, after multiple color passes, this error will gradually expand, resulting in a failed colorized image effect

Dai Zhonghua proposed to use key frame segmentation to segment subsequent frames with the obtained fixed threshold, and use the method of coloring with prior knowledge, which improves the speed. Colorization of subsequent frames is wrong

Images