Dance Video Motion Recognition Based on Computer Vision and Image Processing

Figures & data

Figure 1. Image detection process.

Table 1. Comparison of three filters.

Filter type	Basic principle	Characteristic
Mean filtering	Calculate the average worth of all pixels in the template and replace the central pixel worth of the template with this worth	It can only attenuate the noise, not completely eliminate the noise, but the calculation is simple
Gaussian filtering	When a pixel is weighted, the worth assigned to different positions is different, and the processing effect varies according to the position of the pixel	This method has a good effect on the noise with normal distribution. When the image is smooth, the overall gray distribution characteristics of the image can be preserved
Median filtering	Get the median worth of the template pixel and replace the gray worth of the center pixel with this worth	Median filtering has a good effect on removing salt and pepper noise, and can protect the integrity of image edge information

Figure 2. Extraction process of various motion images.

Table 2. Classification of common features in action recognition.

Category	Form
Static characteristics	Size, color, outline, shape, depth
Dynamic characteristics	Optical flow, speed, speed, direction and track
Spatiotemporal characteristics	Spatiotemporal shape, spatiotemporal interest points, spatiotemporal context
Descriptive features	Attribute, scene, object, posture

Figure 3. Block diagram of layered BP AdaBoost action recognition system.

Figure 4. Effect of average accuracy of different training samples on SOM Network.

Figure 5. Effect of average training time of different training samples on SOM Network.

Table 3. Average accuracy of SOM network under different network size and iteration number combinations.

	Iteration 25 rounds	Iteration 50 rounds	Iteration 75 rounds	Iteration 100 rounds
6×6 network	58.23	64.25	65.78	64.25
8×8 network	57.66	65.44	69.42	70.32
10×10 network	56.13	63.15	66.25	72.58
12×12 network	52.94	61.28	65.12	71.23
15×15 network	51.23	57.69	64.33	68.55

Figure 6. The average accuracy rate under different label noises is based on SVM recognition.

Figure 7. The average training time under different label noises is identified based on SVM.

Figure 8. Average accuracy of BoW-based recognition in different codebook sizes.

Figure 9. Average training time based on BoW recognition in different codebook sizes.