| I&T Solution |
AI Inference Vision of Dump Truck Safety Overseeing System for Worksites
(REF: S-1539) |
| Trial Project |
|
| Solution Feature |
- Dump truck identification: Using object detection algorithm, dump trucks can be recognized and isolated from the rest of the CCTV footage, avoid raising of false alarm non-truck objects.
- Potential over tilting of skip detected: By taking photos of a dump truck with skip raised at different angles, a classification algorithm can be trained to identify the angle a skip has been rase.
- AI inference edge model: does not require much storage space nor computation power to store or execute it, more resources are available for other purposes.
- Alarm system: algorithm can inform safety officers in an obvious way by using siren and throbbing light, easing the burden of safety officers from actively searching for potential overturning events.
- More reliable than manual visual monitoring, as it can operate for long hours and monitor multiple trucks at the same time without impeding performance.
|
| Trial Application and Expected Outcome |
- Dataset containing image of construction sites, or images of dump truck models, is used for testing how well algorithm can locate a dump truck. Should the result be less than desirable, another dataset and/or algorithm would be used to find the most suited one.
- Similar to that of object detection, dataset containing images of dump trucks with skip raised to different angles is used to train and test the effectiveness of classifying the degree. Different datasets, algorithm can be used to compare the suitability of approaches.
- Resources usage can be monitored real-time using prebuild tools within OS, the readings can be recorded and compared with those obtained while running other pretrained models. Storage size of models can be compared against by comparing storage size of file for storing the model.
- Alert flags are sent to visual and audio alarms individually or simultaneously, lux and decibel for brightness and loudness are measured respectively to see if it is sufficient in actual use case. Display would also highlight trucks and have descriptions of each trucks’ status.
- Stress test can be conducted such that the system is made to operate for 48 consecutive hours. To increase loading, multiple CCTV footage can be combined and feed to the monitoring system as one video feed. Results can be reviewed afterwards to determine if the system is robust enough.
|
| Info on I&T Solution Provider |
| Solution Provider | : | SMIT Holdings (HK) Limited | | Address | : | 1/F, Harbour View 2, 16 Science Park East Avenue | | Contact Person | : | Cheung Sin Ming, Ivan |
| Position | : | Technical Director | | Tel | : | 22016321 | | Email | : |
ivancheung@smit.com.cn |
|
