An Innovative SIFT-Based Method for Rigid Video Object Recognition

Algorithm 4

RecognizingthenUpdating(Output:recognized frames, 𝑄; Input: 𝑄, 𝑇𝑄) //to find frames that contain the target object and update feature database with them iteratively.

(1)   𝑇←True //to initiate the loop variable

(2)   While 𝑇 Do //to begin the updating loop

(3)    {

(4)     Set 𝑆𝑄←𝑄, 𝑛←1

(5)     For 𝑖←1 to Num(𝑇𝑄) do   //𝑁𝑢𝑚(𝑇𝑄) represents the number of frame views in 𝑇𝑞.

(6)    {

(7)     Featurepointmatch(Output: 𝑆𝑄, Fsq(𝑓_𝑖); Input: 𝑆𝑄, 𝑓𝑣_𝑖) //to gain matching feature

       keypoint pairs between the feature view 𝑓𝑣_𝑖 and feature models

(8)     Recognizinginfrmae(output: recognized or not, 𝑆(𝑆𝑄), 𝑆(𝐹𝑣_𝑖); input: 𝑆𝑄, Fsq(𝑓_𝑖))

       //to confirm whether the frame image 𝑓_𝑖 contain the target object or not.

(9)     If recognized then

(10)   {

(11)     𝑆𝑆𝑊(𝑛)←𝑓𝑣_𝑖, 𝑛←𝑛+1   //to save the corresponding feature views of the frame image

      into the scalable sliding window, as Figure 6 shown.

(12)     FModelUpdating(Output: 𝑄; Input: 𝑄, 𝑆𝑄, 𝑆(𝑓𝑣_𝑖), Fsq(𝑓𝑣_𝑖)) //to update feature models

      in 𝑄 with recognized feature view 𝑓𝑣_𝑖

(13)   }

(14)  }

(15)     If Empty(𝑆𝑆𝑊) then //to judge the scalable sliding window 𝑆𝑆𝑊 is empty or not

(16)      {

(17)       Break //to jump out the loop

(18)      }

(19)     Else

(20)      {

(21)        For 𝑖←1 to 𝑛−1 do

(22)      {

(23)        For 𝑗←𝑖+1 to 𝑛 do

(24)       {

(25)        FFeatureUpdating(Output: 𝑄; Input: 𝑆𝑄, Fsq(𝑓𝑣_𝑖), Fsq(𝑓𝑣_𝑗))

(26)       }

(27)        Delete(𝑇𝑄, 𝑓𝑣_𝑖) //to delete feature view 𝑓𝑣_𝑖 from the temporal feature database 𝑇𝑄

(28)      }

(29)      }

(30)     Dump(𝑆𝑆𝑊) //to empty 𝑆𝑆𝑊

(31)  }

(32) return 𝑄