The solution for the problem statement resides in solving varied sub problems as discussed below
Classes considered here are Pedestrian,Bi-cycle,Motor-cycle,Car,Truck. (auto-should be added)
custom yolo-v2,with a good fps can be used for Object detection in real time.
Identify detection box for each objecy and compute center y-coordinate of each box.
Box speed calibration is computed by mapping of box-speed in pixels/sec to vehicle-speed in km/hr (miles/Hr for some countries). In order to Create the calibration table we will need a few clips of vehicles and their real speed and for that particular camera setup to get near to the absolute speed.
By determining the flow of detection box, the direction of the vehicle can also be identified
It consist of two parts: (Number Plate Detection + Number Recognition) The Custom Yolo model can take care of the number plate Detection data Pre Processing will help in better recognition of the number and letters.
This solution will help prevent two wheeler fatalities
Custom YOLO method is used. The model only process the Frame Once and all detections can be made. It can therefore be used real time.
This Feature can actually save that few golden minutes that lies in between life and death. Accident dataset is available and this is treated as a classification problem using CNN.Accident recognition can call for an ambulance.
Currently due to the inability of a valid dataset other venues are pursued in search of building an ideal recognition model
the count of the number of Vehicle and pedestrian which can be used to estimate the traffic in real time. This data can be used by the department to optimise the road and traffic
TensorRT(GPU) and OpenVino(CPU) Inference Engine,is to be be used to deploy the model,which will boost the system’s performance .Deepstream can be used to set the pipeline end to end.
We have few bottlenecks with respect to the data availability and Deployment,and the Build For Digital India can act as a perfect platform for us to connect with both Technology experts as well as with the government for the problems implementation.