Method of object consistency detection based on end-to-end deep-learning

Abstract

The invention provides a method of object consistency detection based on end-to-end deep-learning, and aims to simultaneously find a position, a category and consistency of an object in an image. A region-of-interest alignment layer (RoIAlign) is adopted to correctly calculate features of regions of interest (RoIs) from an image feature graph, a convolution layer sequence is utilized to carry outup-sampling on an RoI feature graph to a high-resolution convolution layer to obtain a consistency graph, and a robustness strategy is adopted to adjust a training model to monitor consistency thereof. Object detection is used for object positioning. Consistency detection allocates each pixel in the object to a consistency label thereof, uses multitask loss to carry out training of bounding-box classification, positions and consistency mapping, and finally carries out training and reasoning to obtain consistency labels. According to the method, end-to-end deep-learning is adopted, a multitaskloss function is used to jointly optimize object detection and consistency detection without the need for additional information, complexity in training and testing processes is reduced, and accuracyof detection is effectively improved. The method is suitable for use in application of real-time robots.

Description

Technical field [0001] The invention relates to the field of computer vision, in particular to an object consistency detection method based on end-to-end deep learning. Background technique [0002] In computer vision, detecting and segmenting objects at the same time is becoming more and more popular. Objects can be described by various visual attributes such as color, shape, or physical attributes such as weight, volume, and material. These attributes are useful for identifying objects or classifying them into Different categories are useful. In many robot applications, the consistency of the recognition object is crucial. However, the robot may still need more information to complete the task. The robot must not only detect the consistency of the object, but also be able to locate and recognize Related objects. As an emerging topic, object consistency detection has practical development in many fields, such as scene understanding, video search, object detection, behavior anal...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the problem of time-consuming and unsuitable for real-time applications, the present invention adopts end-to-end deep learning, uses multi-task loss function to jointly optimize object detection and consistency detection, does not require additional information, and reduces the complexity in the training and testing process , which effectively improves the detection accuracy and is suitable for real-time robot applications

Images