An Initial Deep CNN Design Approach for Identification of Vehicle Color and Type for Amber and Silver Alerts

According to the Amber alert report in 2018, of the 132 cases, children in 68 cases were recovered within three or fewer hours from when the amber alert was activated, where 85% involved vehicles. In North Carolina, of the 128 silver alerts, 118 seniors were safely recovered in 2008. Colors, types, and license plate details play a vital role in both Amber and Silver alerts. Currently, children and seniors were recovered when someone recognized the vehicle in the alert. The process can be automated through deep learning modeling to classify a vehicle's color and type with faster detection. The system should also recognize each vehicle's license plate from camera feeds under different weather conditions and find possible matches involved in these emergency alerts to return a child and older adult safely. The automation will help find the number of cases involved in the alert faster and decrease recovery time. The paper focuses on comparing and analyzing the vehicle's color and type classification using Convolution Neural Network (CNN) models for the initial work. Vehicle classification model can be converted to vehicle detection model using Keras, TensorFlow, and OpenCV. Pre-processing includes haze removal, and data augmentation helps to increase testing accuracy and increase the training samples. The best results of each of the models for the color cyan: the precision is 1.00, recall is 0.98, and F1-score is 0.99, and average testing accuracy is 95% type SUV: precision is 0.95, recall is 0.98 and F1-score is 0.96, and average testing accuracy is 89%.