In a recent publication in the journal Sensors, researchers developed a comprehensive safety system incorporating a footrest, data-collection module, and accelerometer for electric scooters.
Study: Revolutionizing Electric Scooter Safety: Innovative Modules and AI Models Mitigate Accidents. Image credit: Generated using DALL.E.3
Background
The electric scooter, a prevalent individual mobility device, transcends gender and age barriers due to its accessibility. S. Gossling contends that micro-mobility, exemplified by electric scooters, can revolutionize transportation. This study addresses the escalating safety concerns associated with the global surge in electric scooter usage.
Despite their popularity, a lack of established traffic laws and safety features exposes riders to accidents. K. Kazemzadeh emphasizes the need for safety research in response to frequent incidents. This global health issue lacks a clear resolution. Comprehensive analyses of electric scooter accidents face challenges due to limited basic data. Studies explore factors contributing to safety risks, including rider behaviors and illegal parking.
T. Brauner identifies common causes, such as riding on sidewalks and multi-person usage, leading to accidents. Various studies propose solutions, from mathematical optimization for parking issues to artificial intelligence (AI) systems monitoring rider concentration. The current study introduces a data-collection module for real-time analysis, offering a preventive approach to electric scooter accidents. The footrest, data collection, and accelerometer modules significantly reduce accident factors and enhance safety.
Enhancing electric scooter safety
In the current study, modules tailored for specific experiments were devised to address the prevalent causes of electric scooter accidents. To combat overloading, a footrest and data collection module were implemented. The footrest incorporated a force-sensitive sensor array, employing Velostat, a force-sensitive material, to gather rider pressure information. This data was transmitted to the data collection module which comprises an analog multiplexer and an Arduino.
The accelerometer module, addressing collisions between scooters and pedestrians or vehicles on sidewalks, integrated an accelerometer and Wi-Fi communication module. Experimental methods involved setting thresholds for the data collection module, eliminating fine wrinkle pressure, and wireless data transmission to a smartphone through the accelerometer module.
Experimental configurations involving footrests and data-collection modules were attached to electric scooters to prevent overloading. Two types of experiments were conducted: one with varying foot positions and the other involving riders and passengers. The accelerometer module, attached to scooters, collected data on paved roads and sidewalks to prevent collisions. Preventing such incidents requires recognizing the scooter's location and distinguishing between sidewalks and paved roads.
The experiment involved collecting accelerometer data during scooter rides on both surfaces, specifically straight, minimally inclined paved roads, over a defined duration and speed, repeating data collection for accuracy. Speed data, collected at a constant speed after reaching a maximum of 10 kilometers per hour, was analyzed to develop and evaluate AI models for each experiment type. The study demonstrates a comprehensive approach to enhancing electric scooter safety, encompassing innovative modules and robust experimental methodologies.
AI model design and evaluation
The setup and evaluation methodology involved the design and training of an AI model tailored to the characteristics of the collected data. The initial model addressed the issue of overloading in electric scooters, utilizing a convolutional neural network (CNN) with softmax and rectified linear unit (ReLU) as activation functions and adaptive moment estimation (ADAM) as an optimizer. The model underwent k-fold cross-validation to ensure reliable accuracy and prevent overfitting.
The second AI model focused on mitigating contact accidents between pedestrians and electric scooters on sidewalks, utilizing an auto-encoder algorithm for regression analysis. This algorithm implemented an anomaly detection method, discerning objects with distinct patterns as anomalies. The evaluation of the second model employed mean absolute error (MAE) and mean square error (MSE) as loss functions, revealing its efficacy in classifying road types based on accelerometer data.
The results showcased the learning convergence processes for both models, demonstrating their capabilities in addressing specific challenges. The MAE loss comparison further validated the learned AI models' effectiveness.
Conclusion
In summary, researchers developed an electric scooter footrest, a data collection module, and an accelerometer module for electric scooter safety. When attached to an electric kickboard, these innovations enable data transmission to a smartphone, connecting to an AI server.
The footrest, made of conductive film, collects boarding data, while the data-collection module, comprising an analog-to-digital converter (ADC) and Arduino module, transmits the analog signals. The accelerometer module captures driving data, transmitting wirelessly through a Wi-Fi module.
Experimentation gathered boarding and driving data, training AI models with CNN and auto-encoder regression. The models exhibited robust accuracy, contributing to the reduction of electric scooter-related accidents. While improvements are recognized, including AI module integration and diverse data collection, the study sets the stage for future research on resolving safety concerns associated with electric scooters.
New Algorithm Boosts Athlete Posture Detection Accuracy