Object Tracking via Frame Segmentation and Motion Vector Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object tracking methods face challenges in achieving real-time tracking while maintaining accuracy, as they are either time-consuming due to frame-by-frame object detection or sacrifice accuracy by not utilizing all image information.
Innovation Solution
The method classifies video frames into detection and tracking types, performing object detection on detection frames and using motion vectors for tracking frames to predict object positions, thereby utilizing all image information for real-time and accurate tracking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If object detection is performed on each image frame, then tracking accuracy is improved, but tracking speed deteriorates
Solution Approach 1:
The patent segments the video processing task by dividing image frames into two types: detection frames (first-type images) and tracking frames (second-type images). Object detection is performed only on detection frames to maintain accuracy, while tracking frames use motion prediction based on motion vectors to improve speed. This segmentation resolves the contradiction by applying different processing strategies to different frame types.
Solution Approach 2:
The patent implements periodic object detection at intervals (on detection frames) rather than continuously on every frame. Between detection frames, tracking is performed using motion prediction. This periodic action maintains tracking accuracy while significantly improving processing speed by reducing the frequency of computationally intensive detection operations.
2Productivity
If object detection is performed on a few extracted images, then tracking speed is improved, but tracking accuracy deteriorates
Solution Approach 1:
The patent performs preliminary object detection on detection frames to obtain accurate position information before tracking. The detected results are then used as initial conditions for motion prediction on subsequent tracking frames. This preliminary action ensures that tracking starts from accurate detected positions, maintaining overall tracking accuracy while enabling faster processing on intermediate frames.
Solution Approach 2:
The patent introduces motion vectors as an intermediary mechanism to bridge detection frames and tracking frames. Motion vectors capture object movement between frames and enable prediction of object positions on tracking frames without requiring full detection. This intermediary approach maintains accuracy by grounding predictions in actual detected positions while achieving speed improvements through reduced computation on tracking frames.
3Measurement precision
If all decoded images are processed for object detection, then tracking accuracy is improved, but time consumption increases
Solution Approach 1:
The patent segments the set of all decoded images into detection frames and tracking frames based on frame type classification. Only detection frames undergo object detection processing, while tracking frames use motion prediction. This segmentation reduces the total number of detection operations from all frames to only detection frames, significantly reducing time consumption while maintaining accuracy through periodic detection.
Solution Approach 2:
The patent implements periodic object detection on detection frames at regular intervals rather than on every decoded image. This periodic approach ensures that detection accuracy is maintained at key intervals while reducing overall detection time by skipping detection on tracking frames, where motion prediction suffices.
Data Source
AI summary
An object tracking method includes: obtaining a video stream acquired by a camera, and decoding the video stream to obtain a plurality of image frames; and obtaining position information of a target object in a current image frame. Obtaining the position information of the target object in the current image frame includes: performing, in response to that the current image frame is a first-type image, object detection on the current image frame to obtain the position information of the target object in the current image frame; and predicting, in response to that the current image frame is a second-type image, the position information of the target object in the current image frame based on motion vectors of a previous image frame and a position the target object in the previous image frame.