Object (spatial) Detection - soil grass - 利用隱藏式馬可夫模型之棒球精彩事件短片偵測

soil grass

4.3 Object (spatial) Detection

The baseball field is characterized by a well-defined layout of specific colors as described in Fig. 4-4. Furthermore, important lines and the bases are in white color, and auditorium (AT) is of high texture and no dominant color as shown in Fig. 4-4(b).

L-AT R-AT

(a) Full view of real baseball field (b) Illustration of baseball field Figure 4-4 The field objects and features.

Each object will be elaborated as follows.

(1) Back auditorium (AT):

The top area which contains high texture and no dominant colors is considered as the auditorium, as the black area above the white horizontal line in Fig. 4-5(a).

(2) Left auditorium (L-AT) and right auditorium (R-AT):

The left area and right area which contains high texture and no dominant colors is considered as the left auditorium and right auditorium, as the left black area and the right black area marked with the white vertical line in Fig. 4-5 (b) and Fig. 4-5 (c).

(a) (b) (c)

Fig. 4-5 Illustration of (a) back auditorium (b) left auditorium (c) right auditorium

(3) Left line (LL) and right line (RL) :

A Ransac algorithm, which finds the line parameter of line segments [12], is applied to the line pixels and then finds the left or right line. The line pixel is high intensity pixel greater than threshold σl excluding pixels in large white area and auditorium area. Either two pixels at a horizontal distance of ±τ pixels or at a vertical distance of

±τ pixels must be darker than σd, where σd << σl. Fig. 4-6 shows the concept of excluding white area. The parameter τ should be set to approximately the double court line width. As illustrated in Fig 4-6, each square represents one pixel and the central one drawn in gray is a candidate pixel. Assuming that white lines are typically no wider than τ pixels (τ = 6 in our system), we check the four pixels, marked ‘V’ and

‘H’, at a distance of τ pixel away from the candidate pixel on the four directions. The central candidate pixel is identified as a white line pixel only if both pixels marked

‘H’ or both pixels marked ‘V’ are with lower brightness than the candidate pixel.

Fig. 4-6 Line pixel detection excluding large white area.

This process prevents that white pixels are extracted in large white areas including auditorium area or white uniforms. Fig. 4-7 is an example of line pixel detection.

Fig. 4-7 (b) shows that the intensity of pixels higher than a threshold of I component in HSI color space and Fig. 4-7 (c) shows that the remaining high intensity pixels after line pixel detection. In Fig. 4-7 (c), the high intensity pixels in pitch mound and

auditorium area is vanished.

Fig. 4-7 The result of retained high intensity pixel after line pixel detection algorithm: (a) original data (b) high intensity pixel before line pixel detection (c) high intensity data after line pixel detection

After line pixel detection, the Ransac algorithm is applied to find line parameter as shown in Fig. 4-8. In Fig 4-8(a), the black point represents the high intensity pixel after the line pixel detection. In Fig 4-8 (b), two points are randomly chosen and the red line is the line passing through the two points. In a specific area (the distance between the point and the line is less than 2, τ = 4), the high intensity pixel will be accumulated denoted a score in this time. The action will repeat until the number of remaining high intensity pixel is less than N or the pre-defined iterative number is reached. Finally, the highest score and the line parameter will be stored as the proposed line in Fig 4-8 (c) and Fig. 4-8 (d).