Human Animation Model Generation in Traffic Accident Restoration: Human Action Recognition Based on Improved DTW Algorithm

Abstract

Traffic accidents are a serious issue in modern society, causing significant personal and property damage. To better understand and prevent these accidents, traffic accident restoration technology has been developed. This study employs a multidimensional approach that includes improved dynamic time warping (DTW) algorithms, dynamic query sets, and Kinect technology. These methods focus on human animation model generation, deformation process design, and human motion recognition. The results demonstrated that the improved DTW algorithm outperformed traditional hidden Markov models, with an average accuracy of 0.92 after 1,000 iterations compared to 0.71. Additionally, the dynamic query set model excelled in computational complexity and time efficiency. The system’s average thigh movement error ranged from 1.6 to 4.0 degrees, with a maximum error of 3.9 to 6.7 degrees and a median error of 1.8 to 3.9 degrees. The improved DTW algorithm maintained better integrity of joint point data in complex motion capture scenarios. The Alpha and Bravo data sets achieved 94.2% and 93.7% accuracy, respectively, significantly higher than the traditional DTW’s 78.8% and 82.3%. The study of human animation model generation in the context of traffic accident restoration has the potential to have a significant impact across a range of fields, enhancing the scientific rigour and practical applicability of traffic accident restoration.

Keywords

• Traffic accidents
• human animation
• dynamic time warping algorithm
• human motion recognition
• dynamic query set