An end-to-end vechicle classification pipeline using vibrometry data

Vehicle Engine Classification Using Spectral Tone-Pitch Vibration Indexing and Neural Network

“…(9) is used instead of DCT of the logarithm of all filter-bank energies in order to capture the possible repeating pattern of the spectral information present in LDV data. Existing temporal and spectral features as tabulated in (Petrovich, Snorrason, & Stevens, 2002) and (Smith, Mendoza-Schrock, Kangas, Derking, & Shaw, 2014), including MFCC have been extremely useful and effective in a great array of speaker and speech recognition applications. However, from our intensive tests using LDV data previous combinations and exhaustive feature selection procedures for vehicle engine classification are far from acceptable (with 20%~50% accuracy rates) and for that reason a new feature targeted to capture LDV data is necessitated.…”

“…Promising performances have been reported thereof. In (Smith, Mendoza-Schrock, Kangas, Derking, & Shaw, 2014) a hierarchical vehicle classification approached using laser-vibrometry data was developed, where a wide array of time and frequency domain features such as spectral flux, MFCC, and number of zero-crossings are tested and automatically selected using a parameter selection procedure to generate a decision tree of different types of vehicles such as vans and sedans. In (Sigmund, Shelley, Bauer, & Heitkamp, 2012) the auto-correlation function of LDV signals was employed as the workhorse to distinguish engine type, speed, and number of cylinders with impressive precision.…”

“…In (Rashed, 2014) the original MFCC after a PCA dimension reduction was fed to a forward neural network to classify different Arabic speakers. In (Petrovich, Snorrason, & Stevens, 2002), vehicle operating conditions were classified using 11 extracted features, including MFCC and others such as zero-crossings, dominant frequencies and Flux, which was also employed in (Smith, Mendoza-Schrock, Kangas, Derking, & Shaw, 2014). …”

Effects of fundamental frequency normalization on vibration-based vehicle classification

“…Linear prediction coefficients (LPC) from directional microphone data were used to train time delay neural networks (TDNN) with 95-100% accuracy for a small set of traffic patterns and vehicle speeds [1] . Features drawn from speech processing, seismic signal processing, and structural analysis were utilized in [2] to identify stationary vehicles with classification rates between 70% and 100% dependent on the classifier.…”

“…Signal processing is necessary for extracting features from the sensing signal for classification. This paper investigates the effects of fundamental frequency normalization on the end-to-end classification process [1] . Using the fundamental frequency, assumed to be the engine's firing frequency, has previously been used successfully to classify vehicles [2,3] .…”