Browse Publications Technical Papers 2019-01-0889

Vision-based techniques for identifying Emergency Vehicles 2019-01-0889

The effectiveness of law enforcement and public safety is directly dependent on the time taken for response by public responders such as police, fire, ambulance, and other public utility vehicles in emergency scenarios. Project RAVEV (Response of Autonomous Vehicles to Emergency Vehicles) investigates how an autonomous vehicle can reliably sense and safely yield to these Emergency Vehicles(EV) on public roads. Safety is imperative here as response maneuvers by the vehicle must not present new risky incidents that lead to collisions. A vehicle can respond to EVs only when it can accurately detect, track and map an EV in its surrounding environment. This paper discusses different computer vision techniques investigated by the authors for identifying EV's. Two independent EV identification frameworks were investigated: (1) A one-stage framework where a custom dataset is trained on an object detection algorithm to detect EV’s, (2) A two-stage framework where different object classification algorithms like SVM, kNN, Adaboost, neural-network-based classifier etc. are employed in series with the KITTI dataset based vehicle detector to classify vehicles into EVs and non-EVs. A comparative study is conducted for different feature vectors derived from multi-spectral or multi-temporal characteristics of an image, along with user-defined feature vectors, which serve as input to classification task in framework 2. Classification outputs from each framework are compared to the ground truth and results are quantitatively listed to conclude upon the ideal decision framework. Once all the EVs in the image frame are identified, a computationally inexpensive object tracking algorithm SORT, which employs Kalman filtering and linear assignment techniques, is used to accurately track EVs in the image frame. This vision-based EV detection scheme fused with multi-sensor data from the vehicle shall be employed to establish a sensor-fusion based EV detection and response framework in a future work.


Subscribers can view annotate, and download all of SAE's content. Learn More »


Attention: This item is not yet published. Pre-Order to be notified, via email, when it becomes available.
Members save up to 40% off list price.
Login to see discount.
Special Offer: With TechSelect, you decide what SAE Technical Papers you need, when you need them, and how much you want to pay.