Reliable Automotive Crash Detection using Multi Sensor Decision Fusion

Due to its critical role in safety of passengers, air bag deployment has to be highly reliable. Existing crash sensing/detection for air bag deployment are based on single sensing device. Multi sensor decision fusion has the potential for increased reliability in decision making using measurements from multiple sensors. In this paper, multi sensor decision fusion algorithm for reliable frontal crash detection and airbag deployment in an automobile using multiple sensors deployed on-board. Decision level fusion of sensor measurement data from an accelerometer located on vehicle engine and a load-cell on the driver shoulder-belt is employed for airbag deployment decision. For state estimation from sensor data, Kalman Filters are developed and tuned for the specific sensors. Crash detection at individual sensor level is also made reliable using a Cumulative Sum Control Chart (CUSUM) based change detection logic from the estimated data. The estimation, detection and decision fusion blocks of the algorithm are implemented in MATLAB and integrated. The resulting crash detection algorithm is tested and validated using US National Highway Traffic Safety Administration (NHTSA) automobile crash data sets. The time taken in the worst case is well within the acceptable bounds of airbag deployment time (50 ms) even with unoptimised MATLAB implementation.