Hybrid Detection System for Automatic Door Activation Using a Time- of-Flight Camera

Abstract

This master’s thesis was conducted in partnership with the BEA Sensors group, a leader company expert in automatic door sensors. The company is located in Liège, Belgium. They design products for opening and safety goals. Nowadays, in a world in constant evolution, it is difficult to stay at the innovation page, especially with the exodus of the artificial intelligence. Every market must adapt the emergence of the new tech- nologies to stay the most competitive. It is also the case in the automatic door systems, the sensors need to consume less energy while staying efficient, providing safety and accuracy. To remain the leader in the automatic doors sensors market, BEA Sensors must diversifies their products and adapt to new technologies. Their interest is to study the capacity of emergent technologies to complete their large products range and always remains the most efficient, consistently providing better products. An emergent technology known as indirect Time-of-Flight camera has attracted their interest. This technology is known for its efficiency in dark and indoor conditions, which can be a crucial benefit for a sensor system. Unfortunately, the Time-of-Flight technology loses in efficiency in the presence of sunlight, which can be an issue if used outdoor. This is a direct consequence of the radiation spectrum of sunlight which has an infrared component. The ToF camera being an infrared light device, the sunlight alters the human detection sensor. Fortunately, the BEA prototype provides an image only constructed using the natural infrared light in the scene, called passive image. This image is better in sunlight conditions than the basic depth images provided by the active measurements. This work provides analyses on the camera’s behavior in sunlight conditions and implements a solution to keep effective human detection and tracking. The proposed solution consists in applying computer vision techniques on the passive component, which measures natural infrared light, to track people in the scene and recover data even when the active component is degraded. The goal is to introduce the passive tracking information in the processing of the active component. This technique is called Sensor Fusion, meaning the fusion between two different measurements information. The results are then evaluated to see if this system could be interesting in the following development of the camera by the company. The evaluation is done on the algorithm efficiency in sunlight conditions, checking if it is able to open doors and keep them open if someone is beneath the camera. The results show that the implemented system is efficient in the corners of the images and in really bright records. However, its efficiency of safety stays limited and the active component remains stable enough to keep the door open.