Use Case Transportation
Linking carton, scan & weights data
As part of its Data & AI transformation, a French transportation and logistics company providing land, sea, and air solutions to and from 157 countries, aims to enhance internal processes, with a focus on improving the performance and productivity of operational teams, particularly in warehouse management and trucking. To accomplish this, it is crucial to link cartons with their respective locations and identifiers.
Challenges
The solution provided for the POC aims to validate the relevance of using data streams, such as video data and scan data, for data matching.
Data quality
The Data quality, especially the camera’s data, weren’t good enough. we needed to perform a lot of pre-processing
Research
Exploration of complex research fields such as gesture recognition
Performances and time
Optimization was necessary for an efficient processing in terms of performances and time
Solution
The solution consists of an application that relies on two data sources: video footage and a corresponding Excel database file. The goal of this solution is first to calculate the dimensions of objects at a warehouse gate and then link these estimated dimensions to the Excel database.
The video footage will be processed through various models and algorithms to extract, as a first step, the dimensions of objects in the warehouse. Next, the person scanning the objects will be detected. This detection will provide data that enables matching between the tabular data (Excel) and the calculated object dimensions.
Tech stack
Results
Since we didn’t have much validation data, only the dimensions estimation part went through validation process. This process involved calculating the mean difference between the estimated dimensions and the actual dimensions of selected objects from the warehouse. Afterward, an SVM model was applied to further enhance the results.
93.65%
Accuracy for dimensions estimation
6.35%
Error for dimensions estimation
Process
The completion of this POC/project required the processing and analysis of three complementary and combinable information sources. Here is an overview of the overall approach.
Key steps carried out to estimate carton dimensions
Diagram of the proposed approach for gesture recognition
Video presentation of the POC demo
