Optical Flow Estimation Using Neural Networks and Egomotion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current motion detection methods using active cameras face challenges in distinguishing target motion from camera motion, leading to false identifications and performance degradation due to noise and video degradation, making real-time analysis difficult.
Innovation Solution
A system and method that utilize state-of-the-art neural networks to determine optical flow fields by incorporating ego-motion of the camera, training the network with image and pose information to ignore camera motion and detect salient objects, applicable to both sparse and dense optical flow.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If current motion detection methods are used with active cameras, then camera motion can be compensated, but false identification of background motion as target occurs and targets are lost in noise
Solution Approach 1:
The patent segments the optical flow estimation into two distinct components: ego-motion flow (camera motion) and object motion flow (target motion). By training separate neural networks to estimate each component independently and then subtracting the ego-motion flow from the total optical flow, the system successfully separates background motion caused by camera movement from actual target motion, eliminating false identifications and preventing targets from being lost in noise.
2Productivity
If state-of-the-art optical flow algorithms are used in presence of video degradation, then motion detection can be performed, but performance greatly diminishes and algorithms become extremely slow
Solution Approach 1:
The patent replaces traditional iterative optical flow algorithms with neural network-based estimation. By training deep learning models on degraded video data, the system achieves robust motion detection performance under various degradation conditions (noise, compression artifacts, reduced frame rates) while maintaining real-time processing speeds, eliminating the performance degradation and computational slowness associated with conventional methods.
3Measurement precision
If neural network is trained with image and pose information, then salient object motion can be detected relative to camera, but system complexity increases
Solution Approach 1:
The patent employs a unified neural network framework that simultaneously processes multiple input types (image data and pose information) to produce optical flow estimation. This multi-functional system handles both ego-motion compensation and object motion detection within a single integrated architecture, achieving high measurement precision for salient object motion while managing system complexity through unified processing rather than separate dedicated systems.
Data Source
AI summary
A camera is connected to a trained neural network. The camera takes an image of a scene and transmits the image to the neural network. A processor connected to the neural network has a localization filter and a robot model implemented therein. A global positioning system (GPS) receiver and inertial measurement unit (IMU) transmit GPS information and IMU information, respectively, to the processor. The localization filter filters the received GPS and IMU information and inputs the filtered information into the robot model. The robot model outputs current position information corresponding to the current image and previous position information corresponding to the respective one or more previous images. The neural network uses the current image and associated current position information and the one or more previous images and respective associated previous position information to generate an estimated optical flow image, which is transmitted to an object detection system.