Alert System for Drivers based on Traffic Signs, Lights and Pedestrian Detection

The research on this topic was important primarily because of the high rate of accidents due to negligence of the traffic signs present. This led to finding ways of creating an alert system for each vehicle which would ensure the safety of those driving the vehicle. Not only should it ensure the safety of the drivers, but also of all the pedestrians and animals through their detection. Many methods were studied including SSD, R-FCN, orthodox neural networks but the implementation of YOLO (You Only Look Once) was carried out because of the efficiency in the image processing rate and the accuracy of the object detection even with custom datasets. The datasets considered include GTSRB, COCO and Bosch for Traffic Signs, Human Detection and Traffic Light detection respectively. The model is trained along with a pre-trained YOLO dataset comprising of 80 classes. The execution results in an accuracy of 91.12% at an image processing speed of 30 fps beating all the methods present including SSD, R-FCN and R-CNN. The above accuracy is the average accuracy obtained on combining the 3 custom datasets, which independently has a much higher accuracy when trained on YOLO.