The physiological studies since the Hubel and Wiesel’s experimentation of cat’s visual cortex have confirmed the consensus about the brain’s intelligence of visual perception. A new way of enhancing the safety of vehicle is proposed by employing the neuromorphic VLSI or processing for mimicking the robust and natural intelligence of visual recognition, inspired by both the Hubel and Wiesel’s experimentation of visual cortex and the neurophysiological model of Hodgkin-Huxley formalism. The feasibility of neuromorphic system is demonstrated successfully for the robust recognition of human objects for the safety either in the car or on the road, evaluating the neuromorphic VLSI implementation based on the controlled CMOS conductance for the bioplausible performance.
The neuromorphic visual information processing is developed for both applications of the driver/occupant analysis in the car and the human object detection on the road. The neuromorphic vision research was motivated by the status analysis of the human posture and safety apparatus for the innovation of the emergency rescue service dealing the crash accidents, and extended its applications to the safety technology of assisting the vehicle drive by detecting nearby pedestrians or human objects. The overall performance is measured with the success rate over 90%, for both the pedestrian detection and the occupant monitoring, in day or night. The most of human object detections are based on the neuromorphic visual information processing using the still image from the video sensor, because of the limited sight condition.
The appropriate use of orientation feature extraction and neural networks ensures the reliability of proposed neuromorphic visual information processing to perform well under various dynamic conditions, such as in the changing ambient light, in night time, or in wet weather which are inevitable for vehicles on the road. The detection of pedestrian or cyclist performs consistently in wide ranges of environment, evaluated in various times and places of Europe and Asia.
The recognition of driver’s eye sight is proved as an added function within the framework of proposed neuromorphic system, to match the varying driver’s eye sight for controlling the eyeglassless 3D dashboard display. The same principle is applicable to detect any particular part or pose of human object, and the neuromorphic visual processing system can accommodate the enforced adaptation or learning as it mimics the natural brain. The neuromorphic coupled with neural networks, suggests it as the new feasible and robust device with the convergence of biological neural system and information technology, or as the cost effective and reliable device of vehicle’s safety enhancement by using the CMOS neuromorphic VLSI approach.