Recognition algorithm for passenger count

3 подписчика

12+
12+

2 просмотра

7 лет назад

ПожаловатьсяНарушение авторских прав

3 подписчика

12+
12+

2 просмотра

7 лет назад

ПожаловатьсяНарушение авторских прав
12+
12+

2 просмотра

7 лет назад

We would like to share with you the results of our continuous works in the field of computer vision. Our team has prepared labeled data and conducted neural network training to create recognition algorithm. Once the machine learned to outline required objects (passengers in this case), we augmented our videostream with labeling the image with 3 different areas: entry (assigned blue color), transit zone and exit (marked in red) Thereby, we specified the count algorithm as follows: passenger was considered to be boarding the bus if he initially appeared at entry area and proceeded to the exit area. Consequently, passenger was considered to be exiting the bus if he moved from exit area to the entry area. We tested the results for count algorithm during the 6-hour videostream. Margin of error was minimal and we achieved 97% accuracy. The team has already begun to tackle the next stage of development, so we are planning to come up with the further steps for this technology.

Название:

Recognition algorithm for passenger count