Research Progress and Challenges of Computer Vision Object Detection Technology

Abstract

With the continuous increase in the number of motor vehicles in the world, traffic accidents caused by drowsy driving have become a major social safety problem. This paper systematically analyzes the evolution path of the current fatigue driving detection technology, focusing on two major methodologies based on the fusion of single features and multiple features. Although the method based on a single feature has the advantages of low computing cost and easy deployment, it is limited by bottlenecks such as environmental interference, device dependence and high false alarm rate. Physiological signal detection can achieve good accuracy in a controlled environment, but it is necessary to solve the problems of individual differences and noise interference. Visual feature detection has significant advantages in non-contact, but the accuracy will be reduced in low-light and occlusion scenes. Although vehicle behavior detection is easy to integrate into the on-board system, the false positive rate is high. The multi-feature fusion method effectively integrates multi-modal data through feature-level, decision-level, and hybrid fusion strategies, which significantly improves the detection accuracy and robustness. In the future, it is necessary to combine lightweight models, edge computing and human-computer collaborative intervention technologies to promote the development of fatigue detection systems in the direction of intelligence and adaptation, so as to provide theoretical support for the realization of all-weather accurate monitoring and traffic safety assurance.