The accuracy and robustness of vehicle localization are critical for achieving safe and reliable high-level autonomy. Recent results show that GPS is vulnerable to spoofing attacks, which is one major threat to autonomous driving. In this paper, a novel anomaly detection and mitigation method against GPS attacks that utilizes onboard camera and high-precision maps is proposed to ensure accurate vehicle localization. First, lateral direction localization in driving lanes is calculated by camera-based lane detection and map matching respectively. Then, a real-time detector for GPS spoofing attack is developed to evaluate the localization data. When the attack is detected, a multi-source fusion-based localization method using Unscented Kalman filter is derived to mitigate GPS attack and improve the localization accuracy. The proposed method is validated in various scenarios in Carla simulator and open-source public dataset to demonstrate its effectiveness in timely GPS attack detection and data recovery.