This paper introduces a new solution for measuring the personal air quality index that reflects the egocentric perspective of human beings with their surrounding environment. Two instances of the solution are introduced and evaluated by using the MediaEval 2019 Insights for wellbeing task dataset and evaluation metric. The first instance calculates the Air Quality Index (AQI) using sensors data, utilizes the user's tags and visual features to measure the personal AQI adaptively. The second instance leverages the average value of the user's tags and feature of the route to determine personal AQI. The performance of these two instances is also discussed.
In [
MediaEval 2019 Insights for wellbeing task [
The definitions, dataset and evaluation metric of this
subtask are described in [
As mentioned above, environmental factors, weather variables, urban nature, and trafic impact on individuals. Observing the dataset provided by the subtask, we found that main streets with lots of trafic and fewer trees will have a low PAQI and vice versa. This observation gives a hint to propose the solution to measure PAQI using AQI, user’s tags, and visual features. Two instances of the solution are introduced and evaluated by using the MediaEval 2019 Insights for wellbeing task dataset and evaluation metric. The first instance calculates the Air Quality Index (AQI) using sensors data, utilizes the user’s tags and visual features to measure the personal AQI adaptively. The second instance leverages the average value of the user’s tags and feature of the route to determine personal AQI. 2.1
First, data along each route are pre-processed to get rid of noises and outliers. Necessary interpolations are conducted to compensate for missing data. Then, two instances (runs) of the proposed solution are constructed as follows: Run 1: From the dataset, we can identify a group of users walking along a specific route. Since the 2018 dataset is recorded by seconds, we convert a recording time to the minute to make sure the highest value of each factor within 1 minute is retained. Then we calculate AQI using these factors (e.g., PM2.5, N O2, O3). Next, visual features are extracted from images.
Run 2: We first collect all data in the same group, then we only keep data coordinating with user’s tags. Next, we divide each segment of one route into four smaller segments. This task aims to have a segment as straight as possible so that the radius can sweep all the points tagged on the segment. Then, we scan the radius with a radius by the distance between the small segments, if any of the tag points are within this range, we collect them and calculate the average value of that user’s tags (e.g., assume the distance between line_start and line_end is 100 m. We divide it into four road segment with 25 m each and get new 3 points in between). 2.2
We use the visual features provided by the task’s organizers. Besides, we develop a tool that crawls images from Google Street View using coordinates provided in the dataset. This tool aims to enrich the image dataset. Finally, we extracted trafic and tree density from these images. 2.3
In this section, we use the data input obtains from section 2.1 corresponding to each run mentioned.
2.3.1 Run 1: PAQI measurement is first calculated by using the AQI calculation formula. Then we use the user’s tags and trafic density, tree density to adjust the AQI values. We build a function to adaptively adjust AQI into PAQI as follows:
Ín f (x ) = i=1(f anctori · αi ) (1) Where: Ín
i=1(αi ) = 1; f actori : input data such as user’s tags and visual feature.
The PAQI’s value are specified by:
PAQI =AQI · f (x ) (2)
Finally, we adjust the value of αi according to a route to get the final PAQI. The value of αi is calculated based on the factors’ values. If the factors’ values are high, αi increases and the PAQI is high. If factors’ values are low, αi decreases and the PAQI is low.
We set parameters as follows: f actor1 ← user’s tags, f actor2 ← trafic density, f actor3 ← tree density, α1 + α2 + α3 = 1. First, we define α1 = α2 = α3 = 31 . Then we use adhoc-based approach to calculate factors’ values and adjust the values of α corresponding with each factor. With f actor1, if its value is larger than the predefined threshold (2.5 in our case), α1 increases, otherwise α1 decreases. With f actor2, if its value is high, α2 decreases, otherwise, α2 increases. With f actor3, if its value is high, α3 increases, otherwise, α3 decreases. This optimal loop is carried on until the convergence happens. With the maximum value of α is 1 and the minimum value of α is 0.
2.3.2 Run 2: First, we based on line_start and line_end points to determine the features of routes 1, 2, and 3 that are featured in Table 1. Second, we calculate the average value of the user’s tags. Third, we calculate the weight of routes:
trail” feature same value with “Bayside path” feature. Table 1 shows the weight of routes 1, 2, and 3 when running on the development dataset. Where: PAQIoutput : is value predicted of routes 4 and 5.; wr : is weight from Table 1.; avд(user ′staдs): the average user’s tags on routes 4 and 5. 3
The experimental results running on the training dataset are denoted in Table 2. The results show we can measure PAQI with acceptable accuracy. Table 3 shows the results when running on the testing dataset. In Table 2 does not include the result of run 2 because we only obtain the weight of routes from Development data and use it to infer PAQI for Testing data. Where: wr : is weight of route.; PAQIintput : based on Development Dataset of routes 1, 2, and 3.; avд(user ′staдs): the average user’s tags on routes 1, 2, and 3.
Because we can not find the “Mountain trail” feature in Development Dataset, we assume the weight of “Mountain (3) 4
In this paper, we report our solution for the challenge raised by MediaEval 2019 Insight for wellbeing task - subtask2. We introduce an ad-hoc approach that adaptively adjust user’s tags, trafic and tree density observed along a route to re-adjust the AQI value towards measuring an acceptable personal AQI value.