Multi-period fusion solution

Multi-period fusion localization is an advanced feature designed to solve the persistence of Mega experiences in complex lighting environments. By building a map set that covers different times, the system can overcome interference to visual features caused by day-night alternation and seasonal changes, ensuring that applications can provide centimeter-level accurate localization at any time of day.

Core challenges

Mega mainly localizes based on visual features in the environment. Although the algorithm has been specifically optimized for lighting and seasonal changes, the dramatic lighting differences caused by day-night alternation can still fundamentally change the visual features of the same place. Therefore, map data collected at only a single time point, such as daytime only, often cannot match in another period, such as nighttime, due to large feature differences, resulting in localization failure.

Solution

To solve all-day localization, the Mega platform provides multi-period data fusion localization capability. By fusing data from different periods in the cloud, it further improves the system's ability to adapt to lighting changes.

How it works

1. Multi-period acquisition: for the same physical scene, collect data under representative different lighting conditions, such as daytime and nighttime.
2. Cloud data fusion: upload all acquired data to Mega Cloud, and the cloud service automatically processes data from these different periods. Through feature fusion optimization, it builds a map database that contains different periods.
3. Automatic matching and localization: when the application runs, the system automatically retrieves and matches the map closest to the current lighting condition from the fused map data according to image features captured by the camera in real time, and returns the pose of the image in the map.

Best practices

To achieve the best fusion effect, follow these acquisition guidelines:

• Cover key periods: include at least one set of "daytime" data and one set of "nighttime" data. For scenes with extremely large lighting changes, such as squares where landscape lights turn on at a fixed time, it is recommended to add acquisitions "before lights on" and "after lights on".
• Path consistency: although acquisition is performed at different times, it is recommended to keep the walking path and shooting angle as consistent as possible for each acquisition. This helps the cloud align and fuse features more efficiently.

Implementation process

To enable multi-period fusion localization, follow a specific acquisition and configuration workflow.

1. Acquisition planning

   Evaluate the lighting changes in the scene and determine the period combinations to acquire

   • Basic combination: one set of daytime data + one set of nighttime data (recommended after streetlights are fully on)
   • Enhanced combination: if there is heavy traffic at dusk and the lighting is special, add another set of dusk data

2. Data acquisition

   When acquiring data for each period, make sure the walking path and shooting angle are as consistent as possible. For example, if the daytime route walks along the centerline of the street from south to north, the nighttime route should keep the same route. This helps the cloud calculate the geometric relationship between maps from different periods more efficiently and greatly improves map alignment accuracy.

   Before starting acquisition for different periods, you need to:

   • Perform acquisition route planning
   • Check GoPro Max configuration or GoPro Max2 configuration
   • Read and become familiar with acquisition method
   • Become familiar with the steps to export GoPro Max/Max2 data

3. Map building

   • If this is your first time using Mega mapping, you need to create a new mapping project
   • In Mega Developer Center, create an ultra-large range multi-map fusion task and upload the acquired multi-period data

4. View mapping results

   After mapping is complete, you can view the mapping results, including acquisition routes and spatial models:

   • Review acquisition routes
   • Preview spatial model

   Tip

   • If mapping fails, refer to mapping failure handling guide
   • If mapping has issues such as layering, refer to model defect handling guide

5. Test localization effect

   • If this is your first time using Mega localization, you need to configure localization map library first
   • Quickly verify whether the localization map library is available
   • View simulation running effect
   • View real onsite running effect

Important

Reminder again: when acquiring data for each period, make sure the walking path and shooting angle are as consistent as possible each time. This helps the cloud calculate spatial relationships between subregions more efficiently and improves map alignment accuracy.

Tip

Multi-period maps are obtained through optimized fusion, and the maps are strictly aligned. Annotation content only needs to be placed once.

Related links

• Device configuration
• Route planning
• Single-map data acquisition requirements
• Large-scale data acquisition
• Mapping failure handling guide
• Model defect handling guide