Intelligent Scene Recognition Based on Deep Learning

Abstract

Using sensor-rich smartphones to sense various contexts attracts much attention, such as transportation mode recognition. Local solutions make efforts to achieve trade-offs among detection accuracy, delay, and battery usage. We propose a real-time recognition model consisting of two long short-term memory classifiers with different sequence lengths. The shorter one is a binary classifier distinguishing elevator scene and the longer one implements a finer classification among bus, subway, high-speed railway, and others. Light-weighted sensors are employed with a much smaller sampling rate (10Hz) compared with previous works. A two-stage setting makes it robust to scenes with different duration and therefore reduces the latency of recognition greatly. Further, the real-time system refines the classification results and attains smoothed predictions. We present experiments on accuracy and resource usage and prove that our system realizes a latency-low and power-efficient scene recognition approach by trading off a reasonable performance loss (averaged recall of 92.22%).

A survey on local transport mode detection on the edge of the network

“…Most local TMD solutions rely on time and frequency domain features to transform sensing information into lower dimensional sets of features [3,8,10]:…”

“…Wang et al [10] collected a data-set of accelerometer, gyroscope, and magnetometer sensors on an Android application. The total duration of the data set is 25.3h.…”

“…In some solutions, position independency has been considered, and such TMD systems have used different ways to show their classifier generalization. For example, some solutions trained the classifier with the data collected from different smartphone positions such as arm, bag, hand, rear/front Yes Yes Yes Martin [20] Yes No No Guvensan [3] Yes Yes Yes Marra [32] N A Y e s N o Soares [38] N A N o N o Liang [22] Y e s N o Y e s Zhao [21] N o N A N A Wang [10] N A N A N o Note: NA=Not Available pocket, chest or waist [3,8,9,20,22] to validate that the classifier generalization is fulfilled; some other solutions used different approaches to limit the side effects of sensor position changes during data collection [9,22]. Moreover, in some systems, user and location independency are examined; for user independency, data from different users with a variety of attributes is collected, and for location independency some approaches have investigated their system generalization by training and testing classifiers with the data coming from different geographical locations.…”

“…In Wangs' system [10] data preprocessing is composed of 3 phases: coordinate transformation (i.e., removing the gravity component from the acceleration on all three axes), Auto-Regressive Moving Average (ARMA) filtering, and data splitting. The ARMA model is a time series model which is usually used to model or compensate for random noise from the sensor readings without introducing a high delay.…”

“…Therefore, there has been a shift towards the use of smartphones for TMD in recent years [7]. The most common solution for smartphone-based TMD is to collect data from smartphones' sensors, train a machine learning (ML) algorithm (i.e., a classifier) based on such data, and then use the generated classifier for making future predictions [3,[8][9][10].…”