AN UNSAFE PROXIMITY DETECTION MODEL TO DECREASE FALSE ALARMS FOR CONSTRUCTION APPLICATIONS
Jun Wang1 and Saiedeh N. Razavi2
1) Ph.D. Candidate, Dept. of Civil Engineering, McMaster University, Hamilton, Canada.
2) Ph.D., Assist. Prof., Dept. of Civil Engineering, McMaster University, Hamilton, Canada.
Abstract: A considerable number of contact collisions happen on construction jobsites due to the continuous and vigorous interactions between various construction entities such as equipment and workers-on-foot. The frequent false alarms generated with current unsafe proximity detection methods have put a limitation on their applicability in real-world environments. In this context, by considering entities’ heading/moving direction and speed along with their position, an unsafe proximity detection model is developed, which has a strong capability in decreasing false alarms. The Extended Kalman Filter (EKF) combined with the Nearest Neighbor (NN) method is developed to track entities’ state information (position, moving direction and speed) in real time. Afterwards, the collected states information is analyzed with the developed safety rules, which are established for five common situations on construction jobsites associated with equipment and workers-on-foot. The safety rules contribute to achieving more accurate unsafe proximity identifications. The unsafe area around equipment is categorized into alert and warning areas and the process to quantify the warning distance is presented using forklift as sample equipment. The model is performed and validated under twelve scenarios in a simulated environment. Four indicators including Root Mean Square Error (RMSE), False Tracking Rate (FTR), False Alarm Rate (FAR) and Model False Rate (MFR) are employed to evaluate model performances. The average FTR is 0.04 indicating an acceptable accuracy of the tracking algorithm. FAR indicates the effectiveness of the safety rules and 66.7% of the obtained FARs are close to 1, which means the model can avoid one false alarm nearly for each scan. Furthermore, diverse potential strategies for implementing the model in real world construction are introduced. The results show that the developed model is promising to promote construction productivity and mobility by reducing interruptions to work in addition to proactively enhancing construction safety.
Keywords: Safety, Unsafe Proximity Detection, State of an Entity, Safety Rules, Quantification of Distance.
Jun Wang and Saiedeh N. Razavi. “AN UNSAFE PROXIMITY DETECTION MODEL TO DECREASE FALSE ALARMS FOR CONSTRUCTION APPLICATIONS.” In Proceedings of International Conference on Civil and Building Engineering Informatics (ICCBEI 2015), 85. Tokyo, Japan, 2015.