Article Preview
Top1. Introduction
Emphasizing the process and neglecting the end is the core idea of the research and implementation of college physical education learning and assessment, while the performance is the main form of assessment results. At present, basketball performance assessment usually consists of two parts, i.e., result assessment and skill assessment. The result assessment mainly includes the number of shots in one minute, half court sports shots, etc. And its assessment process is objective and fair, and the evaluators do not need much professional knowledge. The skill assessment is mainly conducted by professionals according to the skill realization process of testers, which including whether the actions are standard and correct. Compared with the result assessment, the technical assessment is relatively independent and flexible, but it is also very dependent on the professional quality of the evaluators, and is also vulnerable to subjective factors. In addition, the continuation of the COVID-19 epidemic may require Internet teaching and assessment, while online assessment further increases the difficulty of skill assessment. Therefore, it is more and more urgent to research artificial intelligence algorithms that can replace teachers in skill assessment, including feature extraction & feature selection of data analysis (Zheng et al. 2018; Zheng et al. 2022; Zhu et al. 2022), classification of machine learning (Gao et al. 2022; Zhu et al. 2021) etc.
At present, there are many methods for automatic motion recognition, which can be divided into two types: contact type and non-contact type. One branch of the contact type is based on wearable sensors for motion recognition and analysis. However, users using wearable sensors will have a sense of bondage and its cost is very high (Wu et al. 2018). At the same time, this recognition system is vulnerable to external factors to reduce the recognition accuracy. The non-contact motion recognition type mainly uses visual sensors such as cameras to collect data, and uses machine vision and image processing methods to achieve motion recognition and assessment. At present, vision based human motion/posture recognition has become an important topic, and has been widely used in human-computer interaction, virtual reality, intelligent video surveillance and other fields. However, there are still many problems in vision based methods that have not been well solved, which will affect the computer's understanding of human behavior. To be specific, ordinary cameras can only obtain two-dimensional images, but the reconstruction of two-dimensional information to three-dimensional information will lose a lot of important data, affecting the accuracy of motion recognition. Although researchers have designed a variety of image reconstruction algorithms, they are still unable to avoid the effects of lighting, texture occlusion, etc. Kinect sensor uses a new way to obtain images. It captures depth images with spatial distance through a pair of infrared cameras, and extracts skeleton data stream containing three-dimensional coordinate information on the basis of depth images. Therefore, the Kinect body sensor based motion recognition method has gradually attracted extensive attention of researchers.
Kinect is a new generation of somatosensory interactive device that integrates many advanced visual technologies, and it has been highly concerned by many players and researchers in the game and academic circles. Kinect is roughly composed of three parts, RGB camera, depth/infrared camera and infrared sensor, where RGB camera can capture standard two-dimensional color image data stream, and its maximum resolution can reach 1920 × 1080, the speed can reach 30 fps. The depth camera with infrared ray can capture the depth data stream, with a measurement range of 0.5~4.5 m and a resolution of 512 × 424, up to 30 fps. Besides, Kinect can obtain three types of raw data: color image data stream, depth data stream and audio data stream, which correspond to three processing processes: identity recognition, skeleton tracking and speech recognition. However, Kinect does not provide advanced functions for motion/pose recognition, because human movements are so changeable that it is difficult to build a universal model for recognition. In addition, although Kinect sensor can perform basic target object motion capture and speech recognition functions, these functions are low-level and the technology is not perfect. Therefore, it is difficult to directly use Kinect's built-in functions to realize basketball motion recognition and assessment tasks.