Computer vision based loco track visualization

The openCV based loco track visualisation system is a first of its kind implementation in India. The project was done as a part of process excellence in the premise of Madhepura Electric Locomotive Pvt Ltd, a JV between Alstom transport and Indian railways.

Technical blueprint:

Continuous monitoring of locomotives entering the testing building, also known as the carshed, is ensured through 24x7 CCTV camera surveillance. The video frames captured by these cameras are utilized to extract the locomotive ID numbers, which are then directly projected into a customized SCADA software. To minimize errors in this projection process, artificial intelligence (AI) has been extensively implemented. In cases where there is low light in a specific track, the system promptly checks its database to determine if the same locomotive is present in another location. If a match is found, the system rejects projecting the locomotive in the new track, as a locomotive with the same number cannot be present in two different tracks.

Similarly, if the OCR engine reads a garbage value due to sudden locomotive movement or low light conditions, the system scans for the ID numbers of other locomotives in place. During a batch of locomotive production, arbitrary ID numbers are not permitted. Therefore, at any given time, the system calculates the mean of all the locomotive numbers and only allows a variance of +/-20 from the mean number. Considering the production capacity, the plant can produce approximately 10 locomotives within a specific time frame. Hence, the extreme values of the ID numbers cannot deviate more than +/-20 from the mean value. To efficiently handle real-time video streaming from 17 cameras simultaneously, the system incorporates parallel threading. Only after multiple iterations, the locomotive is projected in the visualization system, ensuring a highly reliable and error-proof system.