Detector

The goal of detector is to find turns, line changes, overtakes, obstacle avoidance of vehicle only on the base of mobile phone sensors. The detector should be able to classify those events in real time working on a handset. Most likely the implementation will be a neural network trained with supervised learning. The telemetry of car will be processed on the stage of learning with the above events pointed out manually by a man.

Turn seems to bee the most easy maneuver to detect. The time for which the gyroscope and accelerometer sensors show high values is big enough to easily detect it. Turn detection by means of sensors may be very useful on low speeds or when a car is close to high buildings where gps can give large inaccuracy.

Obstacle avoidance seems to be harder to classify than turns, but still it's a solvable problem. Whenever it occurs a car changes its direction very quickly. Those changes are accompanied with lateral acceleration on the accelerometer sensor and corresponding activity on the gyroscope. Obstacle avoidance may be useful to detect accidents and road potholes. Distributed information about these events can help to automatically set road accidents in navigation software.