Dipak Gaire Sharma

Evolving an emotion recognition module for an intelligent agent using genetic programming and a genetic algorithm

AbstractMost studies use the facial expression to recognize a user’s emotion; however, gestures, such as nodding, shaking the head, or stillness can also be indicators of the user’s emotion. In our research, we use the facial expression and gestures to detect and recognize a user’s emotion. The pervasive Microsoft Kinect sensor captures video data, from which several features representing facial expressions and gestures are extracted. An in-house extensible markup language-based genetic programming engine (XGP) evolves the emotion recognition module of our system. To improve the computational performance of the recognition module, we implemented and compared several approaches, including directed evolution, collaborative filtering via canonical voting, and a genetic algorithm, for an automated voting system. The experimental results indicate that XGP is feasible for evolving emotion classifiers. In addition, the obtained results verify that collaborative filtering improves the generality of recognition. From a psychological viewpoint, the results prove that different people might express their emotions differently, as the emotion classifiers that are evolved for particular users might not be applied successfully to other user(s).

Human gait analysis based on biological motion and evolutionary computing

Abstract Human motion has already deeply affected many aspects of psychological and social research. On the other hand, because of the huge challenges and new dimensions of its increasingly extreme applications, this field remains an inspiring area in which to explore rich possibilities in the fields of artificial intelligence and bio-informatics. In this research, we investigated a novel approach to identify individuals based on their gaits. Furthermore, we investigated a new avenue of the research toward the biometric identification of humans that involves the classification of human gait using the power of genetic programming (GP). Moreover, we also propose an approach that applies collaborative filter using multiple evolved classifiers to address the challenges of non-determinism and insufficient generality of GP.

Human Recognition based on Gait Features and Genetic Programming

Human walking has always been the curious field of research for different disciple of social and information science. The study of human walk or human gait in association with different behaviors and emotions has not only fascinated social science researchers, but its uniqueness has also attracted many computer scientists to work in this arena for the quest of uncovering reliable mechanisms of biometric identification. In this research, we used a novel method for human identification based on inferring the relationship between the human gait features via genetic programming. Moreover, we focus on generating the unique numerical signature that is similar for different locomotion gaits of a particular individual but different across different individuals.

Steering oscillation as an effect of cognitive delay in human drivers

We proposed an approach of applying genetic programming (GP) to automatically develop a driving agent - as a model of a human driver - that optimally steers a realistically simulated car with an instant, non-latent steering response. We verified the hypothesis that introducing a delay in steering response of the evolved model of human driver results in well-expressed steering oscillations. The detection of these oscillations could pave the way for an early-warning of the inadequate cognitive load (as an underlying cause of such delays) of driver in normal driving conditions - well before an urgent response to an eventual hazardous traffic situation is required.

Analysis of genetic programming in gait recognition

Analysis of human motion is one of the most curious fields among different disciplines of socio-psychology, neuro biology and computer science. A persons walk is so crucial because it is associated with lots of other aspects which yield very important information related to emotions, personality, and neurological disorder. The morphological and psychological information induced from both the physiology and neurology involved in motion is one of the key research fields that could one day resolve the various challenges existing in todays intelligent systems inspired from nature. The most interesting among all these is, those wealth of data can be artificially trained using the complex systems in response to generate evidences for identifying the particular person [1]. This paper is the extension of human gait recognition which presents the analysis of efficiency of genetic programming in different cases involved in feature extraction and gait recognition.

Individuality and user-specific approach in adaptive emotion recognition model

This study aims at developing an intelligent agent that can recognize user-specific emotions and can self-evolve. Previous studies have explored several methods to develop the model and improve the results while maintaining the feasibility of real-time implementation for later stages. We evolved the emotion recognition module by using Genetic Programming (GP) and explored several optimizations. We investigated and compared the evolution of a unique classifier (evolved from data from a single specific subject only), the evolution of a general classifier (evolved from data from multiple subjects), and the evolution of an adaptive classifier by implementing incremental GP (evolved incrementally, first from multiple subjects and then from a single specific subject). We conducted the experiments by using the same budget in terms of evolution sessions to obtain the best programs for a fair comparison between general approach, user-specific approach, and adaptive approach. We then performed repeated experiments to verify the robustness of the method. From the results, we concluded that, on an average, adaptive approach not only resulted in faster evolution time, but also achieved better accuracy in emotion recognition.

Evolutionary optimization of weight coefficients of power spectrum for detection of driver-induced steering oscillations

The proposed method applies genetic algorithms (GA) to optimize the weight coefficients of the power spectrum of the Fourier-transformed signal of lateral acceleration of a moving car. The evolved weighted power spectrum detects the steering oscillations caused by the delayed steering response of a human driver in normal, routine driving situations - traffic-less driving on straight and curved roads. Delayed steering response is often a result of drivers having an inadequate cognitive load due to either distraction or cognitive overload. The experimental results, conducted on a realistically simulated car and its environment, indicate that, compared to the power spectrum featuring equal (flat) weight coefficients, the evolved weighted power spectrum facilitates improved discrimination between (i) signals of the lateral acceleration of the car driven by cognitively impaired (i.e., distracted by texting on a mobile phone while driving) drivers and (ii) analog signals of the car, driven by fully attentive drivers. Moreover, for all human drivers who participated in the experiments, the weighted power spectrum of the lateral acceleration of the car driven with distraction, even on a straight section of the road (i.e., inherently featuring lower values of the power spectrum) is even higher than that of the car driven by attentive drivers along curved roads (inherently, featuring higher values of the power spectrum). These results suggest that the proposed method would be applicable for discriminating between subtle driver-induced steering oscillations on straight roads and well-manifested, yet normal steering behavior of drivers when cornering. We view the obtained results as a step towards the development of an early warning system of the inadequate cognitive load of drivers under routine driving conditions - well before any urgent reaction to an eventually dangerous traffic situation might be needed.