Depth information measurement method based on binocular event camera

Abstract

The invention discloses a depth information measurement method based on a binocular event camera, and mainly solves the problem that the existing method is difficult to measure accurate and dense scene depth information. According to the implementation scheme, the method comprises the following steps: constructing a parallax calculation network based on a binocular event camera; acquiring an event stream and a parallax truth value matrix from a data set, performing inductive representation, attention aggregation and binocular matching on the event stream, obtaining a parallax matrix, minimizing the difference between the parallax matrix and the parallax truth value matrix, and realizing training of a parallax calculation network; building a depth information measurement system composed of a binocular camera group, a scene and a computer; and measuring the depth information of an actual scene by using the trained parallax calculation network and the built depth information measurement system. The method can be used to effectively extract the spatio-temporal information and aggregation features of the event stream, so that the method is more robust to noise, the measured dense depth information is more accurate, and the method can be applied to industrial monitoring, medical science, human-computer interaction, virtual reality, reality augmentation and automatic driving.

Description

technical field [0001] The invention belongs to the technical field of measurement, and mainly relates to a method for measuring depth information, which can be applied to industrial monitoring, medical science, human-computer interaction, virtual reality, reality augmentation, and automatic driving scenarios. Background technique [0002] The signals collected by traditional acquisition equipment from the scene only contain two-dimensional information in the spatial domain and do not contain three-dimensional depth information. However, only relying on the two-dimensional information of the scene can no longer meet the needs of actual production and life. How to efficiently obtain 3D depth information from the scene has become a current research hotspot. [0003] Currently, depth information measurement techniques still mainly use standard cameras. Standard cameras generate absolute-intensity image frames at fixed time intervals, which consume a lot of energy and are pron...

AI Technical Summary

Problems solved by technology

[0007] In the Unsupervised event-based learning of optical flow, depth, andegomotion paper, Zhu et al. proposed an unsupervised depth information estimation method based on event cameras. This method performs optical flow estimation and deblurring on events, and maximizes left and right view processing. The similarity of the results to calculate the depth, but this method can only get semi-dense depth results, which is difficult to meet the needs of practical applications.

[0008] Tulyakov et al. proposed a supervised depth information measurement method based on event cameras in the paper Learning an event sequence embedding for dense event-based deep stereo. This method converts events into four-dimensional tensors and uses continuous fully connected networks to extract the time of events. Features, binocular matching is performed on features to calculate depth, but the accuracy of this method to measure depth information still has room for improvement

Images