Information Technology Reference
In-Depth Information
3 Experimental Results
In this section, we evaluate the performance of the proposed object video tracking
method.Wecompareourapproachwithtwootherobjectvideotrackingalgorithms:
structured output tracking with kernels (STRUCK) [1] and tracking-learning-
detection(TLD)method[2].Bothofthealgorithmsareusingtracking-by-detection
approach.
Fig. 4. A snapshot of the test videos used in our experiments. (a) Crossing video. (b)
Diving video. (c) Ironman video.
3.1 Experimental Design
We collect our test videos from TLD [2] and Kwon [18] databases for evaluation,
as shown in Fig. 4. The sequences vary in length from dozens of frames to
hundreds, contain diverse object types (rigid, articulated), have different scene
settings (indoor/outdoor, static/moving camera, lightning conditions). Object
occlusions also present in the test videos. The properties of our test videos are
shown in Table 1. We use Crossing, Diving, and Ironman video as our testing
data. We choose these 3 videos because each video has their own diculties.
 
Search WWH ::




Custom Search