2022年认证杯国际赛（小美赛）C题：Classify Human Activities 思路解析

One important aspect of human behavior understanding is the recognition and monitoring of daily activities. A wearable activity recognition system can improve the quality of life in many critical areas, such as ambulatory monitoring, home-based rehabilitation, and fall detection. Inertial sensor based activity recognition systems are used in monitoring and observation of the elderly remotely by personal alarm systems[1], detection and classification of falls[2], medical diagnosis and treatment[3], monitoring children remotely at home or in school, rehabilitation and physical therapy , biomechanics research, ergonomics, sports science, ballet and dance, animation, film making, TV, live entertainment, virtual reality, and computer games[4]. We try to use miniature inertial sensors and magnetometers positioned on different parts of the body to classify human activities, the following data were obtained.

Each of the 19 activities is performed by eight subjects (4 female, 4 male, between the ages 20 and 30) for 5 minutes. Total signal duration is 5 minutes for each activity of each subject. The subjects are asked to perform the activities in their own style and were not restricted on how the activities should be performed. For this reason, there are inter-subject variations in the speeds and amplitudes of some activities.

Sensor units are calibrated to acquire data at 25 Hz sampling frequency. The 5-min signals are divided into 5-sec segments so that 480(= 60 °¡ 8) signal segments are obtained for each activity.

Your team are asked to develop a reasonable mathematical model to solve the following problems.

1 Please design a set of features and an efficient algorithm in order to classify the 19 types of human actions from the data of these body-worn sensors.

2 Because of the high cost of the data, we need to make the model have a good generalization ability with a limited data set. We need to study and evaluate this problem specifically. Please design a feasible method to evaluate the generalization ability of your model.

3 Please study and overcome the overfitting problem so that your classification algorithm can be widely used on the problem of people’s action classification.