Object identification and classification with ability to return distance-to-object measurement, for use in vehicle collision avoidance system (10 GB 40n6 3HPO)

A small R&D company based in the South East of England is looking for technical partners and/or investment to further develop vehicle collision avoidance system. Potential joint development partners can be from either academia or industry, with experience in the fields of algorithm development and electronics for object identification. Particular emphasis is placed on distance to object measurement.

The company is seeking partners to improve its existing camera based vehicle collision warning system.

The current system's algorithms can (a) identify that the target detected is indeed a vehicle and (b) calculate the distance to that vehicle. This enables the system to warn the driver of hazardous situations. It uses algorithms to identify and calculate the distance to vehicles. The company would like to further develop the system with a view to licensing to Manufacturers/system developers to enable them to achieve a better accuracy of detection rate, when integrated with their existing systems. The company would also like to develop a retrofit version for the aftermarket.

The system demonstrator was developed by the company in collaboration with a well-known Automotive Engineering academic group. The Company has adapted Optical Character Recognition (OCR) code and trained it to return the size of the fonts in the registration plates of vehicles. This, with other control code developed, allows the system to capture distance to target information, which in turn enables the system to positively identify vehicles and issue collision avoidance warnings.

The OCR code only recognises fonts in the image, as opposed to imaging software which returns the values of everything in the image. This greatly reduces processing time. A camera based system typically scans the road ahead taking a series of pictures and each frame has to be processed filtering out all the extraneous data (sky, trees, lampposts road barriers etc.). This company's system only looks for text in that image (the license plate), locks onto it and tracks only that small area of interest, which is a major benefit over existing camera and radar based devices. Also, being an optical system (camera based), it does not need to adhere to the regulatory bandwidth compliance issues of current radar systems. Another problem for radar is cross modulation, as where there are many vehicles fitted with the same system-radar, they interfere with each other. Optical systems do not have this problem.

Detecting the target in front as a valid target is obviously the prime purpose. False returns are a big issue for current systems. A source at Delphi Automotive quotes "36% of the false returns from their radar based system were from non vehicle stationary objects". To overcome these problems, most companies are now developing systems that use a combination of radar and vision "sensor fusion", which increases the cost. In tests, the system developed by this company achieved an accuracy of detection rate of 93%.

Technical Specifications / Specific technical requirements:
The company is seeking partners for the following actions:

(1) Design and implement an algorithm to calculate relative speed between vehicles. Such algorithm will use distance measurement and model-based filtering methods.

(2) Design a system able to detect whether the car in front is the same car as in previous detection, e.g. if, while following a vehicle in front, another car cuts in and encroaches into the vehicle's field of vision, the system detects this by OCR data.

(3) Design an algorithm, which, based on the vehicle model, the results of (1) and (2) and the measured distance between vehicles will suggest an appropriate action to the driver. Possible actions could be: Activate emergency braking, Slow down and then continue with usual distance control.

(4) Detect whether the car in front is in the same lane as the host vehicle i.e. immediately in front, or if not, which lane.

(5) Based on the results of (1), (2), (3) and (4) design an algorithm that will suggest the action to take depending on the detection data returned above, such as:
- Continue behind the car in front at a safe distance.
- Switch to cruise control (there is no traffic immediately ahead).
- Suggest overtaking.

(6) Investigate and design algorithms for stop & go.

(7) Design algorithms for predicting behaviour of surrounding vehicles. Such algorithms will predict future trajectories of the vehicles in front, based on the sequences of past images.

Technologické inovační centrum s.r.o.

Lenka Kostelníková

+420 739 570 792