A Robust Learning-Based Detection and Tracking Algorithm - Technologies and Applications of Artificial Intelligence - page 289

Fig. 5. Tracking results of Crossing sequence: red solid line represents our proposed
method, green dot line represents TLD [2] and blue dashed line represents STRUCK [1]
method. (a) Frame 49. (b) Frame 59. (c) Frame 81.

In the Crossing video, occlusion happens between crossing people and the car
that passes by. While in the Diving video, the diver changes her movement that
caused the target object change its shape. Also, the illumination change happens
in Ironman video.

Tabl e 1. Test Videos Properties

Name Frames Number Target Visible Color Resolution
Crossing 119 119 Yes 360x240

Diving 231 218 Yes 400x224

Ironman 165 165 Yes 720x304

For the experiments, we use 8 pixels sub-blocks size for Crossing video, 16 pix-
els for Diving video and 12 pixels for Ironman video. We use different minimum
sub-blocks size for each video based on the best result from our experimental
results for each video. We use 6 different types of Haar-like features arranged on
a grid at 2 scales on a 4

4 grid, resulting in 192 features, with each features
normalised to give a value in the range [-1,1]. We empploy Gaussian kernel with
˃ =0 . 5. Fig. 5 shows the tracking result of the proposed method and the two
exiting methods when dealing with challenging case.

×