Object Tracking System Using Deep Learning Forecasting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object detection systems using artificial intelligence require significant data computations due to the need for raw video data input, leading to high system performance waste.
Innovation Solution
An object tracking system utilizing a deep learning model to detect an object's area in a video frame, forecast its location in subsequent frames, and determine correlation between areas to track the object, even with dropped frames, reducing data computation by analyzing and predicting object locations in non-analyzed frames.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If raw video data is directly input for object detection, then detection accuracy is maintained, but data computation requirements become too high
Solution Approach 1:
The system performs preliminary actions by detecting objects in only certain video frames (first and second frames) rather than all frames. The deep learning model detects object areas in these selected frames, and then forecast areas are generated for intermediate frames based on the detected positions and motion patterns, reducing the total number of detection operations while maintaining tracking accuracy.
Solution Approach 2:
The system creates forecast areas as simplified copies or predictions of actual object areas in frames where full detection is not performed. These forecast areas are generated by extrapolating from detected object positions in other frames, allowing the system to represent object locations in all frames without performing computationally intensive detection on every frame.
2Reliability
If object detection is performed on every video frame, then continuous tracking is achieved, but system performance is wasted due to excessive computation
Solution Approach 1:
The system performs detection preliminarily in selected frames and uses forecast algorithms to predict object positions in intervening frames. This preliminary detection approach maintains continuous tracking capability while avoiding the excessive computation that would result from detecting every frame, thus improving system performance efficiency.
Solution Approach 2:
Instead of continuous detection on every frame, the system implements periodic detection only on selected frames (first and second frames), with forecast areas generated for intermediate frames. This periodic approach maintains tracking reliability while significantly reducing computational workload and improving overall system efficiency.
3Measurement precision
If deep learning model detects object area in every frame, then tracking precision is improved, but data computation increases significantly
Solution Approach 1:
The deep learning model performs preliminary detection only in selected frames rather than all frames. Forecast areas are generated for intermediate frames based on the detected object areas and motion patterns from the detected frames, maintaining tracking precision while reducing data computation by approximately half.
Data Source
AI summary
An object tracking method includes the following operations: detecting a first area of an object in a first video frame based on a deep learning model, in order to forecast a forecast area of the object in a forecast video frame according to the first video frame and the first area; detecting a second area of the object in a second video frame based on the deep learning model; and determining a correlation between the forecast area and the second area, in order to track the object.