Abstract: The environmental perception system of intelligent driving vehicles is a core module for ensuring safe driving and autonomous decision-making, with target recognition accuracy directly affecting the stability of intelligent control. To address the recognition accuracy challenges faced by new energy commercial vehicles in complex urban environments, this paper constructs a multi-source fusion perception system. It proposes several enhancement methods focusing on multi-sensor fusion strategies, deep learning model optimization, temporal dynamic modeling, and edge deployment. Field tests were conducted in representative urban traffic scenarios. Results show that the optimized model outperforms traditional methods in recognition accuracy, response speed, and small-object detection, verifying its engineering applicability and adaptability. This study provides solid technical support for improving recognition accuracy in intelligent driving systems of new energy commercial vehicles.