Target Detection in Wushu Competition Video Based on Kalman Filter Algorithm of Multi-Target Tracking

Abstract

In a Wushu competition video, target detection for multi-target tracking presents a formidable challenge due to the intricate and dynamic movements inherent in the sport. The task involves identifying and tracking multiple athletes across varying camera angles, lighting conditions, and potentially occluded views. This manuscript present the Dual Transformer Residual Network(DTRN) optimized with Elk Herd Optimizer (EHO) for target detection in wushu competition video of multi-target tracking (WCV-DTRN-EHO). This study utilizes data from the Martial Arts, Dancing and Sports (MADS) and employs pre-processing technique such as Inverse Unscented Kalman Filter (IUKF) based pre-processing process to enhance video quality followed by the Dual Transformer Residual Network (DTRN)is classified as strength, velocity, agility, flexibility, and stamina. The weight parameters of the DTRN are optimized using Elk Herd Optimizer (EHO).The suggested approach is built in Python, and multiple performances evaluation metrics are used to estimate the proposed technique WCV-DTRN-EHO's efficiency for strength, speed, flexibility, agility and endurance in terms of accuracy 26.95%, 28.95%,27.95%,28.95%, and 27.79%, sensitivity 26.38,29.95%,27.95%, 22.13%, and 28.59%, precision 24.87%, 26.83%, 26.95%,24.95%,and 31.23%, F1-score 32.21%, 37.53%, 24.95%,25.95%, and 39.47% and computational time 88.95%, 89.95%,84.35%, 96.47%, and 85.46%while comparing other existing methods such as wushu competition video founded on convolutional neural network (WCV-CNN), wushu competition video based on deep convolutional neural network (WCV-DCNN) and wushu competition video based on recurrent neural network (WCV-RNN) respectively.