Moving Object Detection via Optical Flow Map Warping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional moving object detection techniques in computer vision, particularly those relying on optical flow, face challenges with noise in image and depth sensors, leading to erroneous detections due to irregular foreground object boundaries and undesired holes, which affect the accuracy and computational efficiency of object segmentation.
Innovation Solution
An image processing apparatus and method that captures a sequence of frames, derives optical flow maps to identify motion vectors, warps foreground regions, and applies morphological operations to accurately detect moving objects by combining pixel information across frames, thereby enhancing the robustness of object detection and reducing noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If optical flow procedure is used for moving object detection, then motion information can be extracted from video frames, but noise in the optical flow procedure leads to erroneous detection of objects-of-interest
Solution Approach 1:
The patent combines multiple optical flow maps (first optical flow map from first and second frames, second optical flow map from second and third frames) to generate a combined optical flow map. This merging approach integrates information from multiple frame comparisons, allowing the system to distinguish true motion patterns from noise through consensus across multiple observations, thereby improving both accuracy and reliability of object detection.
Solution Approach 2:
The patent employs a feedback mechanism where the combined optical flow map is used to identify foreground regions, which are then used to generate a refined optical flow map for subsequent processing. This iterative feedback loop allows the system to progressively eliminate noise and erroneous detections by using previously identified foreground information to guide subsequent optical flow calculations, enhancing detection reliability.
2Shape
If foreground detection is performed based on optical flow procedure, then moving objects can be identified, but noise leads to irregular boundaries and undesired holes in foreground regions
Solution Approach 1:
The patent performs preliminary warping of the first foreground region using the first optical flow map before combining it with the second foreground region. This preliminary action of warping aligns the foreground regions in a common coordinate space, ensuring that subsequent combination operations produce smooth, accurate boundaries without irregularities or holes caused by misalignment. The warping step prepares the data in advance for precise merging.
Solution Approach 2:
The patent merges the warped first foreground region with the second foreground region to create a final foreground region. This combining operation integrates multiple detection results, filling in holes and smoothing boundaries by taking the union or intersection of regions detected from different frame pairs, thereby producing a more complete and accurate foreground mask.
3Measurement precision
If multiple optical flow maps are derived and combined, then detection accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent segments the video processing task into distinct stages: generating a first optical flow map from frames 1-2, generating a second optical flow map from frames 2-3, warping the first foreground region, and combining results. This segmentation allows each computational step to be optimized independently and enables parallel processing of different frame pairs, reducing overall computational complexity while maintaining the benefits of multiple optical flow maps.
Solution Approach 2:
The patent performs warping of the first foreground region as a preliminary action before combining it with the second foreground region. By pre-aligning the regions through warping using the first optical flow map, the system avoids the need for more complex real-time alignment algorithms during the combination phase, thereby reducing computational complexity while ensuring accurate merging of foreground regions.
Data Source
AI summary
An apparatus and method for detection of moving objects in a sequence of frames, includes circuitry, a memory, and an image-capture device. The circuitry derives a first optical flow map based on motion information of a first frame and a second frame of a plurality of frames. A second optical flow map is further derived, based on motion information of the second frame and a third frame of the plurality of frames. A first foreground region is further identified that corresponds to the moving objects across the first frame and the second frame and the identified first foreground region is warped across the first frame and the second frame and the moving objects are detected, based on combination of a plurality of pixels in the warped first foreground region and the second foreground region.