The aim is to detect vehicles and track them with individual IDs and estimate the speed.
following pretrained models has been used
- Yolov3 - detection
- DeepSORT - tracking
For speed estimate time stamp based pixel covered is calculated. As a sample mapping is considered (one meter in every 3 pixel) speed of the vehicles is calculated.
To get actual speed one time calibration in terms of pixel to physical distance mapping can be done.
#Note As we move from the bottom to top in image, pixel to distance relation changes exponentially. Equation need to be derived to get approximate speed.
#Some results: