Video Stabilization for Object Recognition in Mobile Platforms
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional object recognition in aerial videos is limited as it can only detect moving targets and is overwhelmed by apparent motion due to platform movement, making it difficult to recognize both moving and stationary objects effectively.
Innovation Solution
A system that stabilizes video frames using image registration and bio-inspired attention algorithms to identify salient points, extract high-dimensional feature vectors, and classify objects, allowing for the detection and recognition of both moving and stationary targets by reducing apparent image motion caused by platform movement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional moving target detection is used in aerial videos, then moving targets can be detected, but stationary targets cannot be detected and detection is overwhelmed by apparent motion due to platform movement
Solution Approach 1:
The system performs video stabilization as a preliminary step before object detection. By stabilizing the video sequence first (removing platform motion effects), the system creates a stable background that enables both moving and stationary target detection to work effectively together in the subsequent processing stages.
Solution Approach 2:
The detection system is divided into separate components: a stable background model (for stationary targets) and a moving target detection component. This segmentation allows each component to specialize in detecting specific target types without interference from platform motion, resolving the contradiction between detecting moving and stationary targets.
2Reliability
If video stabilization is applied to reduce apparent motion, then object recognition performance improves, but computational complexity increases
Solution Approach 1:
Video stabilization is performed as a preliminary preprocessing step before the main object detection and recognition pipeline. This preliminary action removes platform motion effects early, simplifying subsequent detection tasks and improving overall recognition performance despite the added computational step.
Data Source
AI summary
Described is a system for stabilizing, detecting, and recognizing objects in video captured from a mobile platform. The system first receives a video (with a plurality of image frames) captured from a mobile platform. The video is stabilized by registering the image frames to a global coordinate system to generate stabilized image frames. A bio-inspired attention algorithm is applied to the stabilized image frames to produce a set of locations in the stabilized image frames that are salient points representative of an object of interest. An image chip is generated that surrounds each salient point. High-dimensional feature vectors are extracted from the image chip. The feature vectors are then classified as an object class. Thus, through classifying the feature vectors, an object of interest can be identified in the video as captured from the mobile platform.