Multi-period fusion solution

Multi-period fusion positioning is an advanced feature designed to address the persistence of Mega experience in complex lighting environments. By constructing an atlas covering different time periods, the system can overcome the interference of day-night alternation and seasonal changes on visual features, ensuring centimeter-level accurate positioning at any time of the day.

Core challenges

Mega primarily relies on environmental visual features for localization. Although the algorithm has been optimized for factors like lighting and seasonal changes, the drastic illumination differences caused by day-night cycles can still lead to fundamental changes in the visual features of the same location. Therefore, map data collected at a single time point (e.g., only during the day) often fails to match in another time period (e.g., at night) due to significant feature differences, resulting in localization failure.

Solution

To address the issue of all-weather positioning, the Mega platform offers multi-temporal data fusion positioning capability. By performing fusion processing of data from different time periods in the cloud, the system's adaptability to lighting variations is further enhanced.

How it works

1. Multi-session data collection: For the same physical scene, data is collected under representative different lighting conditions (e.g., daytime, nighttime).
2. Cloud-based data fusion: All collected data is uploaded to the Mega cloud, where the cloud service automatically processes these multi-session datasets. Through feature fusion optimization, a map database incorporating different time periods is constructed.
3. Automatic matching and localization: During application runtime, the system automatically retrieves and matches the map closest to the current lighting conditions from the fused map data based on the features captured by the camera in real time, and returns the pose of the image in the map.

Best practices

To achieve optimal fusion results, it is recommended to follow the following collection specifications:

• Cover key time periods: Include at least one set of "daytime" and one set of "nighttime" data. For scenes with extreme lighting changes (such as squares with scheduled landscape lighting), it is advisable to add collections for "before lights-on" and "after lights-on."
• Path consistency: Although collections are taken at different times, it is recommended to maintain consistent walking paths and shooting angles for each collection, as this helps the cloud perform feature alignment and fusion more efficiently.

Implementation process

To enable multi-timeframe fusion positioning, a specific collection and configuration workflow must be followed.

1. Collection planning

   Evaluate the lighting variations in the scene and determine the required combination of timeframes for collection.

   • Basic combination: one set of daytime data + one set of nighttime data (recommended after all streetlights are turned on)
   • Enhanced combination: if the dusk period has high pedestrian traffic and unique lighting, an additional set of dusk data can be added

2. Data collection

   When collecting data for each timeframe, ensure the walking path and camera angles are as consistent as possible. For example: if the daytime route follows the centerline of the street from south to north, the nighttime route should maintain the same path. This helps the cloud more efficiently calculate the geometric relationships between maps of different timeframes, significantly improving map alignment accuracy.

   Before starting data collection for different timeframes, you need to:

   • Perform collection route planning
   • Check GoPro Max device configuration
   • Read and familiarize yourself with the collection method
   • Familiarize yourself with the steps for exporting GoPro Max data

3. Map construction

   • If this is your first time using Mega for mapping, you need to create a new mapping project
   • In the Mega development center, create a large-scale multi-map fusion task and upload the collected multi-timeframe data

4. View mapping results

   After the mapping is completed, you can view the mapping results, including the collection route and spatial model:

   • Review the collection route
   • Preview the spatial model

   Tip

   • If mapping fails, refer to the mapping failure troubleshooting guide for resolution
   • If issues like layering occur during mapping, refer to the model defect handling guide for resolution

5. Test positioning performance

   • If this is your first time using Mega for positioning, first configure the positioning database
   • Quickly verify if the positioning database is usable
   • View the simulation run results
   • View the real-world run results

Important

Reminder: When collecting data for each timeframe, ensure the walking path and camera angles are as consistent as possible. This helps the cloud more efficiently calculate spatial relationships between sub-regions and improves map alignment accuracy.

Tip

Multi-timeframe maps are optimized and fused, ensuring strict alignment between maps. Annotation content only needs to be placed once.

Related links

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