Research On Recognition Method Of Marathon Athlete With Abnormal Behavior Based On Deep Learning

Marathon,as an outdoor long-distance running sport,has attracted more and more people to participate.With the increase in the number of participants,race safety management also faces enormous challenges.Currently,race supervision mostly relies on manually identifying abnormal behaviors through monitoring footage.Due to the limitations of human observation and judgment,negligence or misjudgment often occurs.With the development of artificial intelligence and intelligent monitoring technology,new solutions are provided for race safety management.However,in actual race scenarios,there are multiple actions and dynamic background changes in the video obtained from outdoor monitoring devices.Moreover,capturing moving targets during sports activities results in low-quality situations such as blurriness and deformation in the video,making it difficult to identify abnormal behaviors using existing video-based abnormal behavior recognition methods in practical marathon scenes.This article analyzes the recognition of abnormal behaviors of athletes and athlete recognition based on number plate.Targeted improvements were made to the existing methods to improve the accuracy of abnormal behavior recognition of athletes in marathon scenes.On this basis,a marathon abnormal behavior recognition system was built to improve the efficiency and safety of race supervision.(1)Considering the lack of attention to target area features in existing abnormal behavior recognition methods under motion backgrounds,which leads to the problem of low recognition accuracy of the model,a motion athlete abnormal behavior recognition algorithm based on attention-residual is proposed on the basis of 3D convolutional neural network.Firstly,the composition structure of residual blocks is analyzed on the basis of 3D Res Net-34,and an improved 3D pre-activation residual block is proposed.Secondly,channel and spatio-temporal attention modules are embedded into the residual network to obtain more feature discriminative information and enhance the model’s attention to important features.Finally,experiments are conducted on the athlete behavior dataset.The experimental results show that the proposed method has an increase in recognition accuracy by 7.21% compared to the original algorithm,which further proves the effectiveness of the proposed method.(2)In the identity recognition task of abnormal behavior athletes,a rotation target detection model r Retina Net was proposed to solve the accuracy decline problem caused by the skew distortion of number plate during movement.The model accurately locates the bib number region by introducing angle parameters to match horizontal and rotational targets.After determining the number plate region,the CRNN algorithm was used to recognize the number plate text.The synthetic algorithm was used to generate images of approximate bib number texts for pre-training the network,and real number plate images were used for fine-tuning the pre-trained model.The combination of these two algorithms can effectively solve the problem of abnormal behavior athlete identity recognition.(3)To address the low efficiency and high cost of manually identifying dangerous athlete behavior in current marathon race management,a deep learning-based intelligent recognition system was constructed.The system integrates the athlete abnormal behavior recognition algorithm and the number plate recognition algorithm to identify athlete abnormal behaviors and number plate.The intelligent detection module is the core component of the system,and the video viewing module is used to display the results recognized by the intelligent detection module.The design of this system effectively solves the low efficiency and high cost problems of marathon race safety management.