A Collision Risk Control Model for Mechanical Engineering Vehicles Based on Image Road Condition Monitoring Method

Le résumé fourni par la source

Mechanical engineering vehicles play an important role in various construction projects, but they face various collision risks under different road conditions. To improve the safety of engineering vehicles in operation, a collision risk control model has been proposed. During the process, a generative adversarial network is used as the basis for designing image road condition monitoring methods. DeepLabv3+ is used for generator design, and ResNet101 is used as the encoder backbone network. Subsequently, the risk of vehicle collision is calculated using the artificial potential field method, and the repulsive field adjustment factor is introduced to adjust the repulsive force potential field function. The experiments confirmed that the research method remained below 294 k in parameter testing when the pixel size reached 50 M in two datasets. When testing the obstacle recognition accuracy, the research method achieved a recognition accuracy of 97.8% when obstacles accounted for 10% of the detection area during the day. When conducting obstacle avoidance success rate testing, the obstacle avoidance success rate of the research method remained above 94.7% when the proportion of obstacles in the detection area was 10%. These results confirmed that the research method had good operational performance and collision risk control effect, which could effectively reduce the collision risk of engineering vehicles. The main contribution of this study lies in the proposal of a collision risk control model for mechanical engineering vehicles based on image road condition monitoring and the optimization of the repulsive force potential field function, which solves the problem of local optima.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.