Success of Intelligent Vehicle navigation largely depends on the ability to localize precisely the vehicle in the environment. In general, all intelligent vehicle seemed to agree on a combination of non-cumulative error localization method like GPS with more precise localization method but suffered from cumulative errors like Laser-SLAM with an odometer. However, as GPS is only available for outdoor environment and since the indoor environment is also an important scenario for intelligent vehicles, a replacement of GPS for indoor localization is required. Successfully replacing GPS will not only provide a reliable indoor localization method for vehicles but also keep the architecture of vehicle localizing system consistent and achieve a smooth transition from outdoor to indoor and vice versa. Often, movement speed for indoor vehicles will be as low as 10-12km/h [1] but still, it surpasses the movement speed of human walking (3-5km/h) and presents a challenge for a tight and complex environment. This paper proposes an improved WiFi-fingerprinting method to replace GPS behavior for the indoor environment. The key contribution is to use a raw data smoothing technique with an ensemble classification neural network method to deal with noisy WiFi signal strength. Also, environment constraints are applied to improve localization result. Experiments show this method is capable of replacing GPS for the indoor environment.