Autonomous driving has experienced rapid development in the past decade. Considering that relying solely on onboard sensors limits the safety and efficiency of autonomous driving, introducing a multi-sensory system to perceive road conditions is an effective solution, with roadside units being a typical attempt of a multi-sensory autonomous driving system. However, roadside systems have issues such as high cost, long construction cycles, low coverage, and low short-term utilization rates. Therefore, we propose the Drone-Assisted Multi-Sensory Autonomous Driving System (DAMAD). DAMAD consists of a vehicle perception system, a blind spot detection system, a drone control system, a drone perception system, and an autonomous driving decision system. This paper focuses on the methods of implementing the blind spot identification system and drone sensing system and verifies the performance improvement of DAMAD on DDQN-based deep learning autonomous driving decision systems in a simulation scenario of a typical bidirectional dual-lane multi-intersection. Simulation results show that DAMAD can significantly improve the safety and efficiency of the autonomous driving decision system under ideal conditions. Additionally, we conducted real-vehicle tests, which demonstrated that DAMAD could help autonomous vehicles perceive obscured environmental information in advance, assisting the autonomous driving system in addressing emergencies.