Leaks in water distribution systems are a major problem as a large amount of water is wasted between treatment plants and consumers. Water supply companies use different strategies to identify and locate leaks in their pipe networks, among which, vibro-acoustic methods and devices (i.e. listening sticks, geophones, and leak noise correlators) are the most popular. Although single-point listening devices are cheap and simple to use, they rely on the operator’s hearing training. Moreover, this may lead to a search based on trial and error as only vibration intensity is used, so no information regarding the leak direction is available to aid in its location. Leak noise correlators, however, use two sensors and a leak can be pinpointed by estimating the time of flight between the two measured signals together with knowledge of the speed at which the leak noise propagates. These sensors need to be directly placed on the pipe or a pipe fitting. This paper concerns an investigation of a new technique, where leak noise signals measured using cameras (non-contact measurements), are pre-processed using computer vision techniques to extract meaningful information regarding vibration due to a leak. Here, this is conducted in controlled conditions by using a test bench that simulates the vibration response of a buried pipe at two measurement positions. Tests are carried out to evaluate the camera sensitivity regarding the distance from the signal source, lighting, contrast intensity, and the relation between image resolution and detection of a simulated leak are performed. The results are compared to classic contact measurements made using accelerometers, showing that the proposed technique is promising for leak detection. Moreover, this comparison is conducted using classical signal processing tools such as Power Spectral Density (PSD), modulus and phase of the Cross-Power Spectral Density (CPSD), coherence and Cross-Correlation Coefficient (CCC).