运动科学综述阅读(4): Line-up, Player track

不同于上一节team selection中从atheletes pool中选择小部分人的问题，line-up更多的是order的关系。 所以相关的论文大多采用了模拟(simulation)的方法，也有一些使用GA方法。模拟的问题就在于：开展一次计算是很耗时间的(See Perera et al. 2016).

另一个问题是Player track in sports, 涉及图像识别、定位等问题。在运动分析中，根据运动员的位置、移动情况可以分析运动员的体力情况等数据(See Manafifard et al., 2015)。现有方法多采用PSO，也有使用EA下属的算法的。 使用PSO的目的是在videos中提取运动员，而另一个分析CSA/FA/TLBO的则在一个视频序列(video sequence)中提取运动员。

没有太看懂区别，video sequence和video有什么区别咩… 本综述的作者说分析CSA/FA/TLBO的哪一篇没有提到(poorly mentioned)如何计算适应度

Line-up

creating the best line up of team members within the team
the worst is you do not have data for different lineup

Swartz et al., 2006

SA.The simulations were based on the qualities of the batsmen which were estimated using a Bayesian log linear-model.
fine-tune’s importance
didn’t address the initial team selection, as well as finding optimal bowling lineups

Perera et al. (2016)

focus on optimizing three factor: team selection, batting orders and bowling orders
SA
simulator: Davis et al.(2015) used for the estimation of expected run differential
tuning
considerably larger computational cost on simulations
the only study, to date(2021), that considered all all 3 factors in one(selection of the team, battleing lineup, bowling line up)

Han, 2012

the only in baseball
GA: C1 crossover operator(Reeves, 1996) with swap mutation
lack of similar studies

Player tracking in sports

prediction, object location, object classification,
difficulties: occlusion, fast moving objects, shadows, poor lighting, color contrast, other static background objects
In sports, object detection can be useful in many ways; basically, when analyzing data related to an event
Nakane et al. (2020) influence of meta-heuristics in computer vision

Manafifard et al., 2015

first work, discrete version of PSO
detecting/tracking multiple players and their trajectories in soccer broadcast videos
AdaBoost(a statistical classification meta-algorithm (Freund et al., 1999)).
Solutions: nn matrices(n is the number of players). A matrix represents a particle
velocity also nn matrix
fitness: sum of distances between the nodes in a particle for a given particle
Purpose of PSO algorithm is to identify the trajectories of detected players, which could be used to identify the strengths and weakness of members
PSO is ideal because it includes a velocity component, making the work easier where as in an exhaustive brute search, all edges should bbe explored for the players which will generate huge computaional cost
In summary: used a discrete version of PSO mainly to extract player trajectories

Manafifard et al. (2017)

introduced a method without considering player tracking
blob guided PSO where the PSO is embedded with the blob detection results of the videos
a Blob is a group of connected pixels in an image that share some common property ( E.g gray-scale value)
address the problem of multi-player detection
Players from different teams were identified in each sub swarm. The total search space is divided into sub swarms with respected to the detected blobs and the particles were distributed as uniform as possible within the blobs
In summary: used a modified version of PSO in detecting multiple players in soccer videos

Balaji et al., 2021

volleyball
use meta-heuristics to reduce computational complexity occurred in rael time object detection
In summary: extract a player in a video sequence using three meta-heuristic algorithms