|
|
|
| Untersuchte Arbeit: Seite: 103, Zeilen: 19 |
Quelle: Mairale Gupta 2016 Seite(n): 272, Zeilen: left column, lines 16-18 |
|---|---|
| The Track-Learning-Detect tracker follows an object from frame to frame by localising
all appearances that have been observed so far and corrects the tracker if necessary. The learning estimates detector’s errors and updates it to avoid such errors by using a learning method. The learning process is modelled as a discrete dynamical system and the conditions under which the learning guarantees improvement are found. The downside is that the TLD is computationally intensive. In our testing, we found that when TLD was used, the application would completely crash in older phones due to its high memory consumption. |
The tracker follows the object frame
by frame. The detector stores all appearances that have been observed so far and improves the correctness of tracker if necessary. Learning monitors the performance of both tracker and detector, calculates detector’s errors, and generates training examples to avoid these errors in the future. A novel learning method (P-N learning) which estimates the errors by a pair of “experts”: (i) P-expert estimates missed detections, and (ii) N-expert estimates false alarms. The learning process is modelled as a discrete dynamical system and the conditions under which the learning guarantees improvement are found. |
The phrase "The learning estimates detector’s errors and updates it to avoid such errors by using a learning method" also seems to have been taken from this paper http://vision.stanford.edu/teaching/cs231b_spring1415/papers/Kalal-PAMI.pdf |
|